docs: fix active memory gateway command

Merged via squash.

Prepared head SHA: dfe6a3581c
Co-authored-by: balajisiva <13068516+balajisiva@users.noreply.github.com>
Co-authored-by: scoootscooob <167050519+scoootscooob@users.noreply.github.com>
Reviewed-by: @scoootscooob

.github/labeler.yml

@@ -297,6 +297,10 @@
   - changed-files:
       - any-glob-to-any-file:
           - "extensions/openai/**"
+"extensions: codex":
+  - changed-files:
+      - any-glob-to-any-file:
+          - "extensions/codex/**"
 "extensions: kimi-coding":
   - changed-files:
       - any-glob-to-any-file:

.github/workflows/ci.yml

@@ -37,9 +37,10 @@ jobs:
       run_build_artifacts: ${{ steps.manifest.outputs.run_build_artifacts }}
       run_checks_fast: ${{ steps.manifest.outputs.run_checks_fast }}
       checks_fast_core_matrix: ${{ steps.manifest.outputs.checks_fast_core_matrix }}
-      checks_fast_extensions_matrix: ${{ steps.manifest.outputs.checks_fast_extensions_matrix }}
+      checks_node_extensions_matrix: ${{ steps.manifest.outputs.checks_node_extensions_matrix }}
       run_checks: ${{ steps.manifest.outputs.run_checks }}
       checks_matrix: ${{ steps.manifest.outputs.checks_matrix }}
+      checks_node_core_test_matrix: ${{ steps.manifest.outputs.checks_node_core_test_matrix }}
       run_extension_fast: ${{ steps.manifest.outputs.run_extension_fast }}
       extension_fast_matrix: ${{ steps.manifest.outputs.extension_fast_matrix }}
       run_check: ${{ steps.manifest.outputs.run_check }}
@@ -135,6 +136,9 @@ jobs:
         run: |
           node --input-type=module <<'EOF'
           import { appendFileSync } from "node:fs";
+          import {
+            createNodeTestShards,
+          } from "./scripts/lib/ci-node-test-plan.mjs";
           import {
             createExtensionTestShards,
             DEFAULT_EXTENSION_TEST_SHARD_COUNT,
@@ -211,12 +215,11 @@ jobs:
                   ]
                 : [],
             ),
-            checks_fast_extensions_matrix: extensionShardMatrix,
+            checks_node_extensions_matrix: extensionShardMatrix,
             run_checks: runNode,
             checks_matrix: createMatrix(
               runNode
                 ? [
-                    { check_name: "checks-node-test", runtime: "node", task: "test" },
                     { check_name: "checks-node-channels", runtime: "node", task: "channels" },
                     ...(isPush
                       ? [
@@ -232,6 +235,17 @@ jobs:
                   ]
                 : [],
             ),
+            checks_node_core_test_matrix: createMatrix(
+              runNode
+                ? createNodeTestShards().map((shard) => ({
+                    check_name: shard.checkName,
+                    runtime: "node",
+                    task: "test-shard",
+                    shard_name: shard.shardName,
+                    configs: shard.configs,
+                  }))
+                : [],
+            ),
             run_extension_fast: hasChangedExtensions,
             extension_fast_matrix: createMatrix(
               hasChangedExtensions
@@ -470,7 +484,7 @@ jobs:
               ;;
           esac
 
-  checks-fast-extensions-shard:
+  checks-node-extensions-shard:
     name: ${{ matrix.check_name }}
     needs: [preflight]
     if: needs.preflight.outputs.run_checks_fast == 'true'
@@ -478,7 +492,7 @@ jobs:
     timeout-minutes: 60
     strategy:
       fail-fast: false
-      matrix: ${{ fromJson(needs.preflight.outputs.checks_fast_extensions_matrix) }}
+      matrix: ${{ fromJson(needs.preflight.outputs.checks_node_extensions_matrix) }}
     steps:
       - name: Checkout
         uses: actions/checkout@v6
@@ -497,16 +511,16 @@ jobs:
           OPENCLAW_EXTENSION_BATCH: ${{ matrix.extensions_csv }}
         run: pnpm test:extensions:batch -- "$OPENCLAW_EXTENSION_BATCH"
 
-  checks-fast-extensions:
-    name: checks-fast-extensions
-    needs: [preflight, checks-fast-extensions-shard]
+  checks-node-extensions:
+    name: checks-node-extensions
+    needs: [preflight, checks-node-extensions-shard]
     if: always() && needs.preflight.outputs.run_checks_fast == 'true'
     runs-on: blacksmith-16vcpu-ubuntu-2404
     timeout-minutes: 5
     steps:
       - name: Verify extension shards
         env:
-          SHARD_RESULT: ${{ needs.checks-fast-extensions-shard.result }}
+          SHARD_RESULT: ${{ needs.checks-node-extensions-shard.result }}
         run: |
           if [ "$SHARD_RESULT" != "success" ]; then
             echo "Extension shard checks failed: $SHARD_RESULT" >&2
@@ -599,6 +613,102 @@ jobs:
               ;;
           esac
 
+  checks-node-core-test-shard:
+    name: ${{ matrix.check_name }}
+    needs: [preflight, build-artifacts]
+    if: always() && needs.preflight.outputs.run_checks == 'true' && needs.build-artifacts.result == 'success'
+    runs-on: blacksmith-16vcpu-ubuntu-2404
+    timeout-minutes: 60
+    strategy:
+      fail-fast: false
+      matrix: ${{ fromJson(needs.preflight.outputs.checks_node_core_test_matrix) }}
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v6
+        with:
+          persist-credentials: false
+          submodules: false
+
+      - name: Setup Node environment
+        uses: ./.github/actions/setup-node-env
+        with:
+          node-version: "${{ matrix.node_version || '24.x' }}"
+          cache-key-suffix: "${{ matrix.cache_key_suffix || 'node24' }}"
+          install-bun: "false"
+          use-sticky-disk: "false"
+
+      - name: Configure Node test resources
+        run: echo "OPENCLAW_VITEST_MAX_WORKERS=2" >> "$GITHUB_ENV"
+
+      - name: Download dist artifact
+        uses: actions/download-artifact@v8
+        with:
+          name: dist-build
+          path: dist/
+
+      - name: Download A2UI bundle artifact
+        uses: actions/download-artifact@v8
+        with:
+          name: canvas-a2ui-bundle
+          path: src/canvas-host/a2ui/
+
+      - name: Run Node test shard
+        env:
+          NODE_OPTIONS: --max-old-space-size=6144
+          OPENCLAW_NODE_TEST_CONFIGS_JSON: ${{ toJson(matrix.configs) }}
+        shell: bash
+        run: |
+          set -euo pipefail
+          node --input-type=module <<'EOF'
+          import { spawnSync } from "node:child_process";
+          import { resolveVitestCliEntry, resolveVitestNodeArgs } from "./scripts/run-vitest.mjs";
+
+          const configs = JSON.parse(process.env.OPENCLAW_NODE_TEST_CONFIGS_JSON ?? "[]");
+          if (!Array.isArray(configs) || configs.length === 0) {
+            console.error("Missing node test shard configs");
+            process.exit(1);
+          }
+
+          for (const config of configs) {
+            console.error(`[test] starting ${config}`);
+            const result = spawnSync(
+              "pnpm",
+              [
+                "exec",
+                "node",
+                ...resolveVitestNodeArgs(process.env),
+                resolveVitestCliEntry(),
+                "run",
+                "--config",
+                config,
+              ],
+              {
+                env: process.env,
+                stdio: "inherit",
+              },
+            );
+            if ((result.status ?? 1) !== 0) {
+              process.exit(result.status ?? 1);
+            }
+          }
+          EOF
+
+  checks-node-core-test:
+    name: checks-node-core-test
+    needs: [preflight, checks-node-core-test-shard]
+    if: always() && needs.preflight.outputs.run_checks == 'true'
+    runs-on: blacksmith-16vcpu-ubuntu-2404
+    timeout-minutes: 5
+    steps:
+      - name: Verify node test shards
+        env:
+          SHARD_RESULT: ${{ needs.checks-node-core-test-shard.result }}
+        run: |
+          if [ "$SHARD_RESULT" != "success" ]; then
+            echo "Node test shards failed: $SHARD_RESULT" >&2
+            exit 1
+          fi
+
   extension-fast:
     name: "extension-fast"
     needs: [preflight]

.oxfmtrc.jsonc

@@ -1,9 +1,9 @@
 {
   "$schema": "./node_modules/oxfmt/configuration_schema.json",
-  "experimentalSortImports": {
+  "sortImports": {
     "newlinesBetween": false,
   },
-  "experimentalSortPackageJson": {
+  "sortPackageJson": {
     "sortScripts": true,
   },
   "tabWidth": 2,

.oxlintrc.json

@@ -8,19 +8,23 @@
   },
   "rules": {
     "curly": "error",
-    "eslint-plugin-unicorn/prefer-array-find": "off",
+    "eslint-plugin-unicorn/prefer-array-find": "error",
     "eslint/no-await-in-loop": "off",
-    "eslint/no-new": "off",
+    "eslint/no-new": "error",
     "eslint/no-shadow": "off",
-    "eslint/no-unmodified-loop-condition": "off",
-    "oxc/no-accumulating-spread": "off",
+    "eslint/no-unmodified-loop-condition": "error",
+    "eslint-plugin-unicorn/prefer-set-size": "error",
+    "oxc/no-accumulating-spread": "error",
     "oxc/no-async-endpoint-handlers": "off",
     "oxc/no-map-spread": "off",
+    "typescript/consistent-return": "error",
     "typescript/no-explicit-any": "error",
-    "typescript/no-extraneous-class": "off",
+    "typescript/no-extraneous-class": "error",
+    "typescript/no-unnecessary-type-conversion": "error",
     "typescript/no-unsafe-type-assertion": "off",
     "unicorn/consistent-function-scoping": "off",
-    "unicorn/require-post-message-target-origin": "off"
+    "unicorn/prefer-set-size": "error",
+    "unicorn/require-post-message-target-origin": "error"
   },
   "ignorePatterns": [
     "assets/",
@@ -54,13 +58,7 @@
         "**/*test-support.ts"
       ],
       "rules": {
-        "typescript/await-thenable": "off",
-        "typescript/no-base-to-string": "off",
         "typescript/no-explicit-any": "off",
-        "typescript/no-floating-promises": "off",
-        "typescript/no-misused-spread": "off",
-        "typescript/no-redundant-type-constituents": "off",
-        "typescript/no-unnecessary-template-expression": "off",
         "typescript/unbound-method": "off",
         "eslint/no-unsafe-optional-chaining": "off"
       }

CHANGELOG.md

@@ -6,41 +6,64 @@ Docs: https://docs.openclaw.ai
 
 ### Changes
 
+### Fixes
+
+## 2026.4.10
+
+### Changes
+
+- Models/Codex: add the bundled Codex provider and plugin-owned app-server harness so `codex/gpt-*` models use Codex-managed auth, native threads, model discovery, and compaction while `openai/gpt-*` stays on the normal OpenAI provider path. (#64298)
 - Memory/Active Memory: add a new optional Active Memory plugin that gives OpenClaw a dedicated memory sub-agent right before the main reply, so ongoing chats can automatically pull in relevant preferences, context, and past details without making users remember to manually say "remember this" or "search memory" first. Includes configurable message/recent/full context modes, live `/verbose` inspection, advanced prompt/thinking overrides for tuning, and opt-in transcript persistence for debugging. Docs: https://docs.openclaw.ai/concepts/active-memory. (#63286) Thanks @Takhoffman.
 - macOS/Talk: add an experimental local MLX speech provider for Talk Mode, with explicit provider selection, local utterance playback, interruption handling, and system-voice fallback. (#63539) Thanks @ImLukeF.
+- Tools/video generation: add Seedance 2.0 model refs to the bundled fal provider and submit the provider-specific duration, resolution, audio, and seed metadata fields needed for live Seedance 2.0 runs.
+- Microsoft Teams: add message actions for pin, unpin, read, react, and listing reactions. (#53432) Thanks @sudie-codes.
+- QA/Matrix: add a live `openclaw qa matrix` lane backed by a disposable Matrix homeserver, shared live-transport seams, and Matrix-specific transport coverage for threading, reactions, restart, and allowlist behavior. (#64489) Thanks @gumadeiras.
+- QA/Telegram: add a live `openclaw qa telegram` lane for private-group bot-to-bot checks, harden its artifact handling, and preserve native Telegram command reply threading for QA verification. (#64303) Thanks @obviyus.
+- QA/testing: add a `--runner multipass` lane for `openclaw qa suite` so repo-backed QA scenarios can run inside a disposable Linux VM and write back the usual report, summary, and VM logs. (#63426) Thanks @shakkernerd.
 - CLI/exec policy: add a local `openclaw exec-policy` command with `show`, `preset`, and `set` subcommands for synchronizing requested `tools.exec.*` config with the local exec approvals file, plus follow-up hardening for node-host rejection, rollback safety, and sync conflict detection. (#64050)
 - Gateway: add a `commands.list` RPC so remote gateway clients can discover runtime-native, text, skill, and plugin commands with surface-aware naming and serialized argument metadata. (#62656) Thanks @samzong.
 - Models/providers: add per-provider `models.providers.*.request.allowPrivateNetwork` for trusted self-hosted OpenAI-compatible endpoints, keep the opt-in scoped to model request surfaces, and refresh cached WebSocket managers when request transport overrides change. (#63671) Thanks @qas.
-- QA/testing: add a `--runner multipass` lane for `openclaw qa suite` so repo-backed QA scenarios can run inside a disposable Linux VM and write back the usual report, summary, and VM logs. (#63426) Thanks @shakkernerd.
-- Docs i18n: chunk raw doc translation, reject truncated tagged outputs, avoid ambiguous body-only wrapper unwrapping, and recover from terminated Pi translation sessions without changing the default `openai/gpt-5.4` path. (#62969, #63808) Thanks @hxy91819.
-- Control UI/dreaming: simplify the Scene and Diary surfaces, preserve unknown phase state for partial status payloads, and stabilize waiting-entry recency ordering so Dreaming status and review lists stay clear and deterministic. (#64035) Thanks @davemorin.
-- Gateway: split startup and runtime seams so gateway lifecycle sequencing, reload state, and shutdown behavior stay easier to maintain without changing observed behavior. (#63975) Thanks @gumadeiras.
+- Feishu: standardize request user agents and register the bot as an AI agent so Feishu deployments identify OpenClaw consistently. (#63835) Thanks @evandance.
 - Matrix/partial streaming: add MSC4357 live markers to draft preview sends and edits so supporting Matrix clients can render a live/typewriter animation and stop it when the final edit lands. (#63513) Thanks @TigerInYourDream.
+- Control UI/dreaming: simplify the Scene and Diary surfaces, preserve unknown phase state for partial status payloads, and stabilize waiting-entry recency ordering so Dreaming status and review lists stay clear and deterministic. (#64035) Thanks @davemorin.
+- Agents: add an opt-in strict-agentic embedded Pi execution contract for GPT-5-family runs so plan-only or filler turns keep acting until they hit a real blocker. (#64241) Thanks @100yenadmin.
+- Agents/OpenAI: add provider-owned OpenAI/Codex tool schema compatibility and surface embedded-run replay/liveness state for long-running runs. (#64300) Thanks @100yenadmin.
+- Docs i18n: chunk raw doc translation, reject truncated tagged outputs, avoid ambiguous body-only wrapper unwrapping, and recover from terminated Pi translation sessions without changing the default `openai/gpt-5.4` path. (#62969, #63808) Thanks @hxy91819.
 
 ### Fixes
 
-- fix(nostr): require operator.admin scope for profile mutation routes [AI]. (#63553) Thanks @pgondhi987.
-- Gateway/startup: keep WebSocket RPC available while channels and plugin sidecars start, hold `chat.history` unavailable until startup sidecars finish so synchronous history reads cannot stall startup (reported in #63450), refresh advertised gateway methods after deferred plugin reloads, and enforce the pre-auth WebSocket upgrade budget before the no-handler 503 path so upgrade floods cannot bypass connection limits during that window. (#63480) Thanks @neeravmakwana.
+- Browser/security: tighten browser and sandbox navigation defenses across strict SSRF defaults, hostname allowlists, interaction-driven redirects, subframes, CDP discovery, existing sessions, tab actions, noVNC, marker-span sanitization, and Docker CDP source-range enforcement. (#61404, #63332, #63882, #63885, #63889, #64367, #64370, #64371)
+- Security/tools: harden exec preflight reads, host env denylisting, node output boundaries, outbound host-media reads, profile-mutation authorization, plugin install dependency scanning, ACPX tool hooks, Gmail watcher token redaction, and oversized realtime WebSocket frame handling. (#62333, #62661, #62662, #63277, #63551, #63553, #63886, #63890, #63891, #64459)
+- OpenAI/Codex: add required Codex OAuth scopes, classify provider/runtime failures more clearly, stop suggesting `/elevated full` when auto-approved host exec is unavailable, add OpenAI/Codex tool-schema compatibility, and preserve embedded-run replay/liveness truth across compaction retries and mutating side effects. (#64300, #64439) Thanks @100yenadmin.
+- CLI/WhatsApp media sends: route gateway-mode outbound sends with `--media` through the channel `sendMedia` path and preserve media access context, so WhatsApp document and attachment sends stop silently dropping the file while still delivering the caption. (#64478, #64492) Thanks @ShionEria.
+- Microsoft Teams: restore media downloads for personal DMs, Bot Framework `a:` conversations, OneDrive/SharePoint shared files, and Graph-backed chat IDs; accept Bot Framework audience tokens; prevent feedback-learning filename collisions; keep long tool chains alive with typing indicators; add SSO sign-in callbacks; inject parent context for thread replies; and deliver cron announcements to Teams conversation IDs. (#54932, #55383, #55386, #58001, #58249, #58774, #59731, #60956, #62219, #62674, #63063, #63942, #63945, #63949, #63951, #63953, #64087, #64088, #64089)
 - Gateway/tailscale: start Tailscale exposure and the gateway update check before awaiting channel and plugin sidecar startup so remote operators are not locked out when startup sidecars stall.
+- Gateway/startup: keep WebSocket RPC available while channels and plugin sidecars start, hold `chat.history` unavailable until startup sidecars finish so synchronous history reads cannot stall startup (reported in #63450), refresh advertised gateway methods after deferred plugin reloads, and enforce the pre-auth WebSocket upgrade budget before the no-handler 503 path so upgrade floods cannot bypass connection limits during that window. (#63480) Thanks @neeravmakwana.
 - WhatsApp: keep inbound replies, media, composing indicators, and queued outbound deliveries attached to the current socket across reconnect gaps, including fresh retry-eligible sends after the listener comes back. (#30806, #46299, #62892, #63916) Thanks @mcaxtr.
-- Microsoft Teams: restore media downloads for personal DMs, Bot Framework `a:` conversations, OneDrive/SharePoint shared files, and Graph-backed chat IDs; accept Bot Framework audience tokens; and deliver cron announcements to Teams conversation IDs. (#55383, #58001, #58249, #62219, #62674, #63063, #63942, #63951, #63953) Thanks @obviyus.
-- Gateway/thread routing: preserve Slack, Telegram, Mattermost, and ACP parent-thread delivery targets so subagent, cron, and stream-relay completion messages land back in the originating thread or topic. (#54840, #57056, #63228, #63506) Thanks @yzzymt.
+- Gateway/thread routing: preserve Slack, Telegram, Mattermost, Matrix, ACP, restart-sentinel, and agent announce delivery targets so subagent, cron, stream-relay, session fallback, and restart messages land back in the originating thread, topic, or room casing. (#54840, #57056, #63143, #63228, #63506, #64343, #64391)
+- Models/fallback: preserve `/models` selection across transient primary-model failures and config reloads, allow timeout cooldown probes, classify OpenRouter no-endpoints responses, detect llama.cpp context overflows, and keep provider/runtime context metadata stable through reloads. (#61472, #64196, #64471)
+- Agents/BTW: keep `/btw` side questions working after tool-use turns by stripping replayed tool blocks, hidden reasoning, and malformed image payloads, omitting empty tool arrays, allowing Bedrock `auth: "aws-sdk"`, and routing Feishu `/btw` plus `/stop` through bounded out-of-band lanes. (#64218, #64219, #64225, #64324) Thanks @ngutman.
+- Control UI/BTW: render `/btw` side results as dismissible ephemeral cards in the browser, send `/btw` immediately during active runs, and clear stale BTW cards on reset flows so webchat matches the intended detached side-question behavior. (#64290) Thanks @ngutman.
+- Commands/targeting: use the selected agent or session for command output, send policy, usage/cost, context reports, model lists, bash sandbox hints, BTW/compact working directories, plugin commands, and session exports so multi-agent commands describe and mutate the intended target instead of the requester.
+- Conversation bindings: normalize focused/current conversation ids, preserve binding metadata on account and Discord rebinds, avoid stale Discord lifecycle windows, and keep generic activity touches persisted so reply routing survives rebinds and restarts.
+- iMessage/self-chat: distinguish normal DM outbound rows from true self-chat using `destination_caller_id` plus chat participants, preserve multi-handle self-chat aliases, drop ambiguous reflected echoes, and strip wrapped imsg RPC text fields. (#61619, #63868, #63980, #63989, #64000) Thanks @neeravmakwana.
+- Matrix: keep multi-account room scoping consistent, keep packaged crypto migrations warning-only when appropriate, preserve ordered block streaming, add explicit Matrix block-streaming opt-in, and resolve verification/bootstrap from the packaged runtime entry. (#58449, #59249, #59266, #64373) Thanks @gumadeiras.
+- Telegram/security: tighten Telegram `allowFrom` sender validation and keep `/whoami` allowlist reporting in sync with command auth checks.
 - Agents/timeouts: extend the default LLM idle window to 120s and keep silent no-token idle timeouts on recovery paths, so slow models can retry or fall back before users see an error.
 - Gateway/agents: preserve configured model selection and richer `IDENTITY.md` content across agent create/update flows and workspace moves, and fail safely instead of silently overwriting unreadable identity files. (#61577) Thanks @samzong.
+- Skills/TaskFlow: restore valid frontmatter fences for the bundled `taskflow` and `taskflow-inbox-triage` skills and copy bundled `SKILL.md` files as hard dist-runtime copies so skills stay discoverable and loadable after updates. (#64166, #64469) Thanks @extrasmall0.
+- Skills: respect overridden home directories when loading personal skills so service, test, and custom launch environments read the intended user skill directory instead of the process home.
 - Windows/exec: settle supervisor waits from child exit state after stdout and stderr drain even when `close` never arrives, so CLI commands stop hanging or dying with forced `SIGKILL` on Windows. (#64072) Thanks @obviyus.
 - Browser/sandbox: prevent sandbox browser CDP startup hangs by recreating containers when the browser security hash changes and by waiting on the correct sandbox browser lifecycle. (#62873) Thanks @Syysean.
-- iMessage/self-chat: distinguish normal DM outbound rows from true self-chat using `destination_caller_id` plus chat participants, while preserving multi-handle self-chat aliases so outbound DM replies stop looping back as inbound messages. (#61619) Thanks @neeravmakwana.
 - QQBot/streaming: make block streaming configurable per QQ bot account via `streaming.mode` (`"partial"` | `"off"`, default `"partial"`) instead of hardcoding it off, so responses can be delivered incrementally. (#63746)
 - QQBot/config: allow extra fields in `channels.qqbot` and `channels.qqbot.accounts.*` so extended qqbot builds can add new config options without gateway startup failing on schema validation. (#64075) Thanks @WideLee.
 - Dreaming/gateway: require `operator.admin` for persistent `/dreaming on|off` changes and treat missing gateway client scopes as unprivileged instead of silently allowing config writes. (#63872) Thanks @mbelinky.
-- Matrix/multi-account: keep room-level `account` scoping, inherited room overrides, and implicit account selection consistent across top-level default auth, named accounts, and cached-credential env setups. (#58449) thanks @Daanvdplas and @gumadeiras.
 - Gateway/pairing: prefer explicit QR bootstrap auth over earlier Tailscale auth classification so iOS `/pair qr` silent bootstrap pairing does not fall through to `pairing required`. (#59232) Thanks @ngutman.
 - Browser/control: auto-generate browser-control auth tokens for `none` and `trusted-proxy` modes, and route browser auth/profile/doctor helpers through the public browser plugin facades. (#63280, #63957) Thanks @pgondhi987.
 - Browser/act: centralize `/act` request normalization and execution dispatch while adding stable machine-readable route-level error codes for invalid requests, selector misuse, evaluate-disabled gating, target mismatch, and existing-session unsupported actions. (#63977) Thanks @joshavant.
-- Security/exec: replace script-preflight check-then-read logic with an atomic pinned-file-descriptor open, and expand the host environment denylist for dangerous runtime-control variables. (#62333, #63277) Thanks @pgondhi987.
-- Security/nodes: keep `nodes` tool output paths inside the workspace boundary so model-driven node writes cannot escape the intended workspace. (#63551) Thanks @pgondhi987.
 - Security/QQBot: enforce media storage boundaries for all outbound local file paths and route image-size probes through SSRF-guarded media fetching instead of raw `fetch()`. (#63271, #63495) Thanks @pgondhi987.
 - Channel setup: ignore workspace plugin shadows when resolving trusted channel setup catalog entries so onboarding and setup flows keep using the bundled, trusted setup contract.
+- Gateway/memory startup: load the explicitly selected memory-slot plugin during gateway startup, while keeping restrictive allowlists and implicit default memory slots from auto-starting unrelated memory plugins. (#64423) Thanks @EronFan.
 - Config/plugins: let config writes keep disabled plugin entries without forcing required plugin config schemas or crashing raw plugin validation, and avoid re-activating plugin registry state during schema checks. (#54971, #63296) Thanks @fuller-stack-dev.
 - Config validation: surface the actual offending field for strict-schema union failures in bindings, including top-level unexpected keys on the matching ACP branch. (#40841) Thanks @Hollychou924.
 - Wizard/plugin config: coerce integer-typed plugin config fields from interactive text input so integer schema values persist as numbers instead of failing validation. (#63346) Thanks @jalehman.
@@ -58,56 +81,56 @@ Docs: https://docs.openclaw.ai
 - Claude CLI: clear inherited Anthropic auth/header environment aliases before spawning Claude Code and add sanitized CLI backend auth-env diagnostics for debugging gateway-run provider selection.
 - Agents/failover: classify AbortError and stream-abort messages as timeout so Ollama NDJSON stream aborts stop showing `reason=unknown` in model fallback logs. (#58324) Thanks @yelog.
 - Fireworks/FirePass: disable Kimi K2.5 Turbo reasoning output by forcing thinking off on the FirePass path and hardening the provider wrapper so hidden reasoning no longer leaks into visible replies. (#63607) Thanks @frankekn.
-- Matrix/multi-account: keep room-level `account` scoping, inherited room overrides, and implicit account selection consistent across top-level default auth, named accounts, and cached-credential env setups. (#58449) Thanks @gumadeiras.
-- Matrix/runtime: resolve the verification/bootstrap runtime from a distinct packaged Matrix entry so global npm installs stop failing on crypto bootstrap with missing-module or recursive runtime alias errors. (#59249) Thanks @gumadeiras.
-- Matrix/streaming: preserve ordered block flushes before tool, message, and agent boundaries, add explicit `channels.matrix.blockStreaming` opt-in so Matrix `streaming: "off"` stays final-only by default, and move MiniMax plain-text final handling into the MiniMax provider runtime instead of the shared core heuristic. (#59266) Thanks @gumadeiras.
-- QQBot/streaming: make block streaming configurable per QQ bot account via `streaming.mode` (`"partial"` | `"off"`, default `"partial"`) instead of hardcoding it off, so responses can be delivered incrementally. (#63746)
 - Discord: update Carbon to v0.15.0. Thanks @thewilloftheshadow.
 - Config/Discord: coerce safe integer numeric Discord IDs to strings during config validation, keep unsafe or precision-losing numeric snowflakes rejected, and align `openclaw doctor` repair guidance with the same fail-closed behavior. (#45125) Thanks @moliendocode.
 - BlueBubbles/config: accept `enrichGroupParticipantsFromContacts` in the core strict config schema so gateways no longer fail validation or startup when the BlueBubbles plugin writes that field. (#56889) Thanks @zqchris.
 - Feishu/webhooks: read webhook bodies through the pre-auth guard so unauthenticated webhook traffic stays under the same body budget as other protected channel ingress paths.
 - Tools/web_fetch: add an opt-in `tools.web.fetch.ssrfPolicy.allowRfc2544BenchmarkRange` config so fake-IP proxy environments that resolve public sites into `198.18.0.0/15` can use `web_fetch` without weakening the default SSRF block. (#61830) Thanks @xing-xing-coder.
-- Dreaming/gateway: require `operator.admin` for persistent `/dreaming on|off` changes and treat missing gateway client scopes as unprivileged instead of silently allowing config writes. (#63872) Thanks @mbelinky.
 - Dreaming/cron: reconcile managed dreaming cron from startup config and runtime lifecycle changes, but only recover managed dreaming cron state during heartbeat-triggered dreaming checks so ordinary chat traffic does not recreate removed jobs. (#63873, #63929, #63938) Thanks @mbelinky.
 - Memory/lancedb: accept `dreaming` config when `memory-lancedb` owns the memory slot so Dreaming surfaces can read slot-owner settings without schema rejection. (#63874) Thanks @mbelinky.
 - Control UI/dreaming: keep the Dreaming trace area contained and scrollable so overlays no longer cover tabs or blow out the page layout. (#63875) Thanks @mbelinky.
+- Dreaming/narrative: harden request-scoped diary fallback so scheduled dreaming only falls back on the dedicated subagent-runtime error, stop trusting spoofable raw error-code objects, and avoid leaking workspace paths when local fallback writes fail. (#64156) Thanks @mbelinky.
 - Dreaming/diary: add idempotent narrative subagent runs, preserve restrictive `DREAMS.md` permissions during atomic writes, and surface temp cleanup failures so repeated sweeps do not double-run the same narrative request or silently weaken diary safety. (#63876) Thanks @mbelinky.
 - Heartbeats/sessions: remove stale accumulated isolated heartbeat session keys when the next tick converges them back to the canonical sibling, so repaired sessions stop showing orphaned `:heartbeat:heartbeat` variants in session listings. (#59606) Thanks @rogerdigital.
 - Gateway/run cleanup: fix stale run-context TTL cleanup so the new maintenance sweep resets orphaned run sequence state and prevents unbounded run-context growth. (#52731) Thanks @artwalker.
 - UI/compaction: keep the compaction indicator in a retry-pending state until the run actually finishes, so the UI does not show `Context compacted` before compaction actually finishes. (#55132) Thanks @mpz4life.
 - Cron/tool schemas: keep cron tool schemas strict-model-friendly while still preserving `failureAlert=false`, nullable `agentId`/`sessionKey`, and flattened add/update recovery for the newly exposed cron job fields. (#55043) Thanks @brunolorente.
 - Git metadata: read commit ids from packed refs as well as loose refs so version and status metadata stay accurate after repository maintenance. (#63943)
-- Gateway: keep `commands.list` skill entries categorized under tools and include provider-aware plugin `nativeName` metadata even when `scope=text`, so remote clients can group skills correctly and map text-surface plugin commands back to native aliases.
+- Gateway: keep `commands.list` skill entries categorized under tools and include provider-aware plugin `nativeName` metadata even when `scope=text`, so remote clients can group skills correctly and map text-surface plugin commands back to native aliases. (#64147)
 - TUI: reset footer activity to idle when switching sessions so a stale streaming indicator cannot persist after the selection changes. (#63988) Thanks @neeravmakwana.
-- iMessage: treat `sender === chat_identifier` as self-chat only when `destination_caller_id` is present and matches the sender, fixing DM outbound rows that omit destination from being run through self-chat echo handling. (#63980) Thanks @neeravmakwana.
-- Cron/Telegram: collapse isolated announce delivery to the final assistant-visible text only for Telegram targets, while preserving existing multi-message direct delivery semantics for other channels. (#63228) Thanks @welfo-beo.
-- Gateway/thread routing: preserve Slack, Telegram, and Mattermost thread-child delivery targets so bound subagent completion messages land in the originating thread instead of top-level channels. (#54840) Thanks @yzzymt.
-- ACP/stream relay: pass parent delivery context to ACP stream relay system events so `streamTo="parent"` updates route to the correct thread or topic instead of falling back to the main DM. (#57056) Thanks @pingren.
-- Agents/sessions: preserve announce `threadId` when `sessions.list` fallback rehydrates agent-to-agent announce targets so final announce messages stay in the originating thread/topic. (#63506) Thanks @SnowSky1.
-- iMessage/self-chat: remember ambiguous `sender === chat_identifier` outbound rows with missing `destination_caller_id` in self-chat dedupe state so the later reflected inbound copy still drops instead of re-entering inbound handling when the echo cache misses. Thanks @neeravmakwana.
 - Claude CLI: stop marking spawned Claude Code runs as host-managed so they keep using normal CLI subscription behavior. (#64023) Thanks @Alex-Alaniz.
-- Agents/failover: classify OpenRouter `404 No endpoints found for <model>` responses as `model_not_found` so fallback chains continue past retired OpenRouter candidates. (#61472) Thanks @MonkeyLeeT.
-- Browser/plugin SDK: route browser auth, profile, host-inspection, and doctor readiness helpers through browser plugin public facades so core compatibility helpers stop carrying duplicate runtime implementations. (#63957) Thanks @joshavant.
-- Agents/failover: allow cooldown probes for `timeout` (including network outage classifications) so the primary model can recover after failover without a gateway restart. (#63996) Thanks @neeravmakwana.
-- iMessage (imsg): strip an accidental protobuf length-delimited UTF-8 field wrapper from inbound `text` and `reply_to_text` when it fully consumes the field, fixing leading garbage before the real message. (#63868) Thanks @neeravmakwana.
+- Codex auth: brand Codex OAuth flows as OpenClaw in user-visible auth prompts and diagnostics.
 - Gateway/pairing: fail closed for paired device records that have no device tokens, and reject pairing approvals whose requested scopes do not match the requested device roles.
 - ACP/gateway chat: classify lifecycle errors before forwarding them to ACP clients so refusals use ACP's refusal stop reason while transient backend errors continue to finish as normal turns.
-- Agents/BTW: strip replayed tool blocks, hidden reasoning, and malformed image payloads from `/btw` side-question context so Bedrock no-tools side questions keep working after tool-use turns. (#64225) Thanks @ngutman.
-- Commands/btw: keep tool-less side questions from sending injected empty `tools` arrays on strict OpenAI-compatible providers, so `/btw` continues working after prior tool-call history. (#64219) Thanks @ngutman.
-- Agents/Bedrock: let `/btw` side questions use `auth: "aws-sdk"` without a static API key so Bedrock IAM and instance-role sessions stop failing before the side question runs. (#64218) Thanks @SnowSky1.
-- Feishu: route `/btw` side questions and `/stop` onto bounded out-of-band lanes so BTW no longer waits behind a busy normal chat turn while ordinary same-chat traffic stays FIFO. (#64324) Thanks @ngutman.
-- Agents/failover: detect llama.cpp slot context overflows as context-overflow errors so compaction can retry self-hosted OpenAI-compatible runs instead of surfacing the raw upstream 400. (#64196) Thanks @alexander-applyinnovations.
 - Claude CLI/skills: pass eligible OpenClaw skills into CLI runs, including native Claude Code skill resolution via a temporary plugin plus per-run skill env/API key injection. (#62686, #62723) Thanks @zomars.
 - Discord: keep generated auto-thread names working with reasoning models by giving title generation enough output budget for thinking plus visible title text. (#64172) Thanks @hanamizuki.
-- Heartbeat: ignore doc-only Markdown fence markers in the default `HEARTBEAT.md` template so comment-only heartbeat scaffolds skip API calls again. (#63434) Thanks @ravyg.
-- Control UI/BTW: render `/btw` side results as dismissible ephemeral cards in the browser, send `/btw` immediately during active runs, and clear stale BTW cards on reset flows so webchat matches the intended detached side-question behavior. (#64290) Thanks @ngutman.
+- Heartbeat: ignore doc-only Markdown fence markers in the default `HEARTBEAT.md` template so comment-only heartbeat scaffolds skip API calls again. (#61690, #63434) Thanks @ravyg.
 - Reply/skills: keep resolved skill and memory secret config stable through embedded reply runs so raw SecretRefs in secondary skill settings no longer crash replies when the gateway already has the live env. (#64249) Thanks @mbelinky.
-- Dreaming/startup: keep plugin-registered startup hooks alive across workspace hook reloads and include dreaming startup owners in the gateway startup plugin scope, so managed Dreaming cron registration comes back reliably after gateway boot. (#62327) Thanks @mbelinky.
+- Dreaming/startup: keep plugin-registered startup hooks alive across workspace hook reloads and include dreaming startup owners in the gateway startup plugin scope, so managed Dreaming cron registration comes back reliably after gateway boot. (#62327, #64258) Thanks @mbelinky.
 - Plugins: treat duplicate `registerService` calls from the same plugin id as idempotent so snapshot and activation loads no longer emit spurious `service already registered` diagnostics. (#62033, #64128) Thanks @ly85206559.
 - Discord/TTS: route auto voice replies through the native voice-note path so Discord receives Opus voice messages instead of regular audio attachments. (#64096) Thanks @LiuHuaize.
-- Config/plugins: use plugin-owned command alias metadata when `plugins.allow` contains runtime command names like `dreaming`, and point users at the owning plugin instead of stale plugin-not-found guidance. (#64242) Thanks @feiskyer.
+- Config/plugins: use plugin-owned command alias metadata when `plugins.allow` contains runtime command names like `dreaming`, and point users at the owning plugin instead of stale plugin-not-found guidance. (#64191, #64242) Thanks @feiskyer.
 - Agents/Gemini: strip orphaned `required` entries from Gemini tool schemas so provider validation no longer rejects tools after schema cleanup or union flattening. (#64284) Thanks @xxxxxmax.
-- Assistant text: strip Qwen-style XML tool call payloads from visible replies so web and channel messages no longer show raw `<tool_call><function=...>` output. (#64214) Thanks @MoerAI.
+- Assistant text: strip Qwen-style XML tool call payloads from visible replies so web and channel messages no longer show raw `<tool_call><function=...>` output. (#63999, #64214) Thanks @MoerAI.
+- Daemon/gateway: prevent systemd restart storms on configuration errors by exiting with `EX_CONFIG` and adding generated unit restart-prevention guards. (#63913) Thanks @neo1027144-creator.
+- Agents/exec: prevent gateway crash ("Agent listener invoked outside active run") when a subagent exec tool produces stdout/stderr after the agent run has ended or been aborted. (#62821) Thanks @openperf.
+- Gateway/OpenAI compat: return real `usage` for non-stream `/v1/chat/completions` responses, emit the final usage chunk when `stream_options.include_usage=true`, and bound usage-gated stream finalization after lifecycle end. (#62986) Thanks @Lellansin.
+- Agents/subagents: deduplicate delivered completion announces so retry or re-entry cleanup does not inject duplicate internal-context completion turns into the parent session. (#61525) Thanks @100yenadmin.
+- Agents/exec: keep sandboxed `tools.exec.host=auto` sessions from honoring per-call `host=node` or `host=gateway` overrides while a sandbox runtime is active, and stop advertising node routing in that state so exec stays on the sandbox host. (#63880)
+- Agents/subagents: preserve archived delete-mode runs until `sessions.delete` succeeds and prevent overlapping archive sweeps from duplicating in-flight cleanup attempts. (#61801) Thanks @100yenadmin.
+- Cron/isolated agent: run scheduled agent turns as non-owner senders so owner-only tools stay unavailable during cron execution. (#63878)
+- Discord/sandbox: include `image` in sandbox media param normalization so Discord event cover images cannot bypass sandbox path rewriting. (#64377) Thanks @mmaps.
+- Agents/exec: extend exec completion detection to cover local background exec formats so the owner-downgrade fires correctly for all exec paths. (#64376) Thanks @mmaps.
+- Hooks/security: mark agent hook system events as untrusted and sanitize hook display names before cron metadata reuse. (#64372) Thanks @eleqtrizit.
+- Daemon/launchd: keep `openclaw gateway stop` persistent without uninstalling the macOS LaunchAgent, re-enable it on explicit restart or repair, and harden launchd label handling. (#64447) Thanks @ngutman.
+- Plugins/context engines: preserve `plugins.slots.contextEngine` through normalization and keep explicitly selected workspace context-engine plugins enabled, so loader diagnostics and plugin activation stop dropping that slot selection. (#64192) Thanks @hclsys.
+- Heartbeat: stop top-level `interval:` and `prompt:` fields outside the `tasks:` block from bleeding into the last parsed heartbeat task. (#64488) Thanks @Rahulkumar070.
+- Agents/OpenAI replay: preserve malformed function-call arguments in stored assistant history, avoid double-encoding preserved raw strings on replay, and coerce replayed string args back to objects at Anthropic and Google provider boundaries. (#61956) Thanks @100yenadmin.
+- Heartbeat/config: accept and honor `agents.defaults.heartbeat.timeoutSeconds` and per-agent heartbeat timeout overrides for heartbeat agent turns. (#64491) Thanks @cedillarack.
+- CLI/devices: make implicit `openclaw devices approve` selection preview-only and require approving the exact request ID, preventing latest-request races during device pairing. (#64160) Thanks @coygeek.
+- Media/security: honor sender-scoped `toolsBySender` policy for outbound host-media reads so denied senders cannot trigger host file disclosure via attachment hydration. (#64459) Thanks @eleqtrizit.
+- Browser/security: reject strict-policy hostname navigation unless the hostname is an explicit allowlist exception or IP literal, and route CDP HTTP discovery through the pinned SSRF fetch path. (#64367) Thanks @eleqtrizit.
+- Models/vLLM: ignore empty `tool_calls` arrays from reasoning-model OpenAI-compatible replies, reset false `toolUse` stop reasons when no actual tool calls were parsed, and stop sending `tool_choice` unless tools are present so vLLM reasoning responses no longer hang indefinitely. (#61197, #61534) Thanks @balajisiva.
 
 ## 2026.4.9
 
@@ -156,7 +179,6 @@ Docs: https://docs.openclaw.ai
 - Control UI/models: preserve provider-qualified refs for OpenRouter catalog models whose ids already contain slashes so picker selections submit allowlist-compatible model refs instead of dropping the `openrouter/` prefix. (#63416) Thanks @sallyom.
 - Plugin SDK/command auth: split command status builders onto the lightweight `openclaw/plugin-sdk/command-status` subpath while preserving deprecated `command-auth` compatibility exports, so auth-only plugin imports no longer pull status/context warmup into CLI onboarding paths. (#63174) Thanks @hxy91819.
 - Wizard/plugin config: coerce integer-typed plugin config fields from interactive text input so integer schema values persist as numbers instead of failing validation. (#63346) Thanks @jalehman.
-- Dreaming/narrative: harden request-scoped diary fallback so scheduled dreaming only falls back on the dedicated subagent-runtime error, stop trusting spoofable raw error-code objects, and avoid leaking workspace paths when local fallback writes fail. (#64156) Thanks @mbelinky.
 
 ## 2026.4.8
 
@@ -277,6 +299,9 @@ Docs: https://docs.openclaw.ai
 - CLI/tasks: `openclaw tasks cancel` now records operator cancellation for CLI runtime tasks instead of returning "Task runtime does not support cancellation yet", so stuck `running` CLI tasks can be cleared. (#62419) Thanks @neeravmakwana.
 - Sessions/context: resolve context window limits using the active provider plus model (not bare model id alone) when persisting session usage, applying inline directives, and sizing memory-flush / preflight compaction thresholds, so duplicate model ids across providers no longer leak the wrong `contextTokens` into the session store or `/status`. (#62472) Thanks @neeravmakwana.
 - Channels/setup: exclude workspace shadow entries from channel setup catalog lookups and align trust checks with auto-enable so workspace-scoped overrides no longer bypass the trusted catalog. (`GHSA-82qx-6vj7-p8m2`) Thanks @zsxsoft.
+- Reply execution: prefer the active runtime snapshot over stale queued reply config during embedded reply and follow-up execution so SecretRef-backed reply turns stop crashing after secrets have already resolved. (#62693) Thanks @mbelinky.
+- Android/manual connect: allow blank port input only for TLS manual gateway endpoints so standard HTTPS Tailscale hosts default to `443` without silently changing cleartext manual connects. (#63134) Thanks @Tyler-RNG.
+- Matrix/agents: hide owner-only `set-profile` from embedded agent channel-action discovery so non-owner runs stop advertising profile updates they cannot execute. (#62662) Thanks @eleqtrizit.
 
 ## 2026.4.5
 
@@ -496,7 +521,7 @@ Docs: https://docs.openclaw.ai
 - Agents/scheduling: steer background-now work toward automatic completion wake and treat `process` polling as on-demand inspection or intervention instead of default completion handling. (#60877) Thanks @vincentkoc.
 - Agents/skills: skip `.git` and `node_modules` when mirroring skills into sandbox workspaces so read-only sandboxes do not copy repo history or dependency trees. (#61090) Thanks @joelnishanth.
 - ACP/agents: inherit the target agent workspace for cross-agent ACP spawns and fall back safely when the inherited workspace no longer exists. (#58438) Thanks @zssggle-rgb.
-- ACPX/Windows: preserve backslashes and absolute `.exe` paths in Claude CLI parsing, and fail fast on wrapper-script targets with guidance to use `cmd.exe /c`, `powershell.exe -File`, or `node <script>`. (#60689) Thanks @steipete.
+- ACPX/Windows: preserve backslashes and absolute `.exe` paths in Claude CLI parsing, and fail fast on wrapper-script targets with guidance to use `cmd.exe /c`, `powershell.exe -File`, or `node <script>`. (#60689)
 - Auth/failover: persist selected fallback overrides before retrying, shorten `auth_permanent` lockouts, and refresh websocket/shared-auth sessions only when real auth changes occur so retries and secret rotations behave predictably. (#60404, #60323, #60387) Thanks @extrasmall0 and @mappel-nv.
 - Gateway/channels: pin the initial startup channel registry before later plugin-registry churn so configured channels stay visible and `channels.status` stops falling back to empty `channelOrder` / `channels` payloads after runtime plugin loads.
 - Prompt caching: order stable workspace project-context files before `HEARTBEAT.md` and keep `HEARTBEAT.md` below the system-prompt cache boundary so heartbeat churn does not invalidate the stable project-context prefix. (#58979) Thanks @yozu and @vincentkoc.
@@ -1467,7 +1492,7 @@ Docs: https://docs.openclaw.ai
 - Gateway/status: tolerate network interface discovery failures in status, onboarding control-UI links, and self-presence display paths so those surfaces fall back cleanly instead of crashing. (#52195) Thanks @meng-clb.
 - Gateway/Linux: auto-detect nvm-managed Node TLS CA bundle needs before CLI startup and refresh installed services that are missing `NODE_EXTRA_CA_CERTS`. (#51146) Thanks @GodsBoy.
 - Google auth/Node 25: patch `gaxios` to use native fetch without injecting `globalThis.window`, while translating proxy and mTLS transport settings so Google Vertex and Google Chat auth keep working on Node 25. (#47914) Thanks @pdd-cli.
-- Gateway/plugins: pin runtime webhook routes to the gateway startup registry so channel webhooks keep working across plugin-registry churn, and make plugin auth + dispatch resolve routes from the same live HTTP-route registry. (#47902) Fixes #46924 and #47041. Thanks @steipete.
+- Gateway/plugins: pin runtime webhook routes to the gateway startup registry so channel webhooks keep working across plugin-registry churn, and make plugin auth + dispatch resolve routes from the same live HTTP-route registry. (#47902) Fixes #46924 and #47041.
 - Gateway/restart: defer externally signaled unmanaged restarts through the in-process idle drain, and preserve the restored subagent run as remap fallback during orphan recovery so resumed sessions do not duplicate work. (#47719) Thanks @joeykrug.
 - Telegram/setup: seed fresh setups with `channels.telegram.groups["*"].requireMention=true` so new bots stay mention-gated in groups unless you explicitly open them up. Thanks @vincentkoc.
 - Inbound policy hardening: tighten callback and webhook sender checks across Mattermost and Google Chat, match Nextcloud Talk rooms by stable room token, and treat explicit empty Twitch allowlists as deny-all. (#46787) Thanks @zpbrent, @ijxpwastaken and @vincentkoc.
@@ -2880,7 +2905,7 @@ Docs: https://docs.openclaw.ai
 - Gemini OAuth/Auth flow: align OAuth project discovery metadata and endpoint fallback handling for Gemini CLI auth, including fallback coverage for environment-provided project IDs. (#16684) Thanks @vincentkoc.
 - Google Chat/Lifecycle: keep Google Chat `startAccount` pending until abort in webhook mode so startup is no longer interpreted as immediate exit, preventing auto-restart loops and webhook-target churn. (#27384) thanks @junsuwhy.
 - Temp dirs/Linux umask: force `0700` permissions after temp-dir creation and self-heal existing writable temp dirs before trust checks so `umask 0002` installs no longer crash-loop on startup. Landed from contributor PR #27860. (#27853) Thanks @stakeswky.
-- Nextcloud Talk/Lifecycle: keep `startAccount` pending until abort and stop the webhook monitor on shutdown, preventing `EADDRINUSE` restart loops when the gateway manages account lifecycle. (#27897) Thanks @steipete.
+- Nextcloud Talk/Lifecycle: keep `startAccount` pending until abort and stop the webhook monitor on shutdown, preventing `EADDRINUSE` restart loops when the gateway manages account lifecycle. (#27897)
 - Microsoft Teams/File uploads: acknowledge `fileConsent/invoke` immediately (`invokeResponse` before upload + file card send) so Teams no longer shows false "Something went wrong" timeout banners while upload completion continues asynchronously; includes updated async regression coverage. Landed from contributor PR #27641 by @scz2011.
 - Queue/Drain/Cron reliability: harden lane draining with guaranteed `draining` flag reset on synchronous pump failures, reject new queue enqueues during gateway restart drain windows (instead of silently killing accepted tasks), add `/stop` queued-backlog cutoff metadata with stale-message skipping (while avoiding cross-session native-stop cutoff bleed), and raise isolated cron `agentTurn` outer safety timeout to avoid false 10-minute timeout races against longer agent session timeouts. (#27407, #27332, #27427)
 - Typing/Main reply pipeline: always mark dispatch idle in `agent-runner` finalization so typing cleanup runs even when dispatcher `onIdle` does not fire, preventing stuck typing indicators after run completion. (#27250) Thanks @Sid-Qin.
@@ -2897,7 +2922,7 @@ Docs: https://docs.openclaw.ai
 - Agents/Canvas default node resolution: when multiple connected canvas-capable nodes exist and no single `mac-*` candidate is selected, default to the first connected candidate instead of failing with `node required` for implicit-node canvas tool calls. Landed from contributor PR #27444. Thanks @carbaj03.
 - TUI/stream assembly: preserve streamed text across real tool-boundary drops without keeping stale streamed text when non-text blocks appear only in the final payload. Landed from contributor PR #27711 by @scz2011. (#27674)
 - Hooks/Internal `message:sent`: forward `sessionKey` on outbound sends from agent delivery, cron isolated delivery, gateway receipt acks, heartbeat sends, session-maintenance warnings, and restart-sentinel recovery so internal `message:sent` hooks consistently dispatch with session context, including `openclaw agent --deliver` runs resumed via `--session-id` (without explicit `--session-key`). Landed from contributor PR #27584. Thanks @qualiobra.
-- Pi image-token usage: stop re-injecting history image blocks each turn, process image references from the current prompt only, and prune already-answered user-image blocks in stored history to prevent runaway token growth. (#27602) Thanks @steipete.
+- Pi image-token usage: stop re-injecting history image blocks each turn, process image references from the current prompt only, and prune already-answered user-image blocks in stored history to prevent runaway token growth. (#27602)
 - BlueBubbles/SSRF: auto-allowlist the configured `serverUrl` hostname for attachment fetches so localhost/private-IP BlueBubbles setups are no longer false-blocked by default SSRF checks. Landed from contributor PR #27648 by @lailoo. (#27599) Thanks @taylorhou for reporting.
 - Agents/Compaction + onboarding safety: prevent destructive double-compaction by stripping stale assistant usage around compaction boundaries, skipping post-compaction custom metadata writes in the same attempt, and cancelling safeguard compaction when there are no real conversation messages to summarize; harden workspace/bootstrap detection for memory-backed workspaces; and change `openclaw onboard --reset` default scope to `config+creds+sessions` (workspace deletion now requires `--reset-scope full`). (#26458, #27314) Thanks @jaden-clovervnd, @Sid-Qin, and @widingmarcus-cyber for fix direction in #26502, #26529, and #27492.
 - NO_REPLY suppression: suppress `NO_REPLY` before Slack API send and in sub-agent announce completion flow so sentinel text no longer leaks into user channels. Landed from contributor PRs #27529 (by @Sid-Qin) and #27535 (rewritten minimal landing by maintainers). (#27387, #27531)
@@ -2919,7 +2944,7 @@ Docs: https://docs.openclaw.ai
 - LINE/Inline directives auth: gate directive parsing (`/model`, `/think`, `/verbose`, `/reasoning`, `/queue`) on resolved authorization (`command.isAuthorizedSender`) so `commands.allowFrom`-authorized LINE senders are not silently stripped when raw `CommandAuthorized` is unset. Landed from contributor PR #27248 by @kevinWangSheng. (#27240)
 - Onboarding/Gateway: seed default Control UI `allowedOrigins` for non-loopback binds during onboarding (`localhost`/`127.0.0.1` plus custom bind host) so fresh non-loopback setups do not fail startup due to missing origin policy. (#26157) thanks @stakeswky.
 - Docker/GCP onboarding: reduce first-build OOM risk by capping Node heap during `pnpm install`, reuse existing gateway token during `docker-setup.sh` reruns so `.env` stays aligned with config, auto-bootstrap Control UI allowed origins for non-loopback Docker binds, and add GCP docs guidance for tokenized dashboard links + pairing recovery commands. (#26253) Thanks @pandego.
-- CLI/Gateway `--force` in non-root Docker: recover from `lsof` permission failures (`EACCES`/`EPERM`) by falling back to `fuser` kill + probe-based port checks, so `openclaw gateway --force` works for default container `node` user flows. (#27941) Thanks @steipete.
+- CLI/Gateway `--force` in non-root Docker: recover from `lsof` permission failures (`EACCES`/`EPERM`) by falling back to `fuser` kill + probe-based port checks, so `openclaw gateway --force` works for default container `node` user flows. (#27941)
 - Gateway/Bind visibility: emit a startup warning when binding to non-loopback addresses so operators get explicit exposure guidance in runtime logs. (#25397) thanks @let5sne.
 - Sessions cleanup/Doctor: add `openclaw sessions cleanup --fix-missing` to prune store entries whose transcript files are missing, including doctor guidance and CLI coverage. Landed from contributor PR #27508 by @Sid-Qin. (#27422)
 - Doctor/State integrity: ignore metadata-only slash routing sessions when checking recent missing transcripts so `openclaw doctor` no longer reports false-positive transcript-missing warnings for `*:slash:*` keys. (#27375) thanks @gumadeiras.
@@ -2981,24 +3006,24 @@ Docs: https://docs.openclaw.ai
 - Slack/Threading: stop forcing tool-call reply mode to `all` based on `ThreadLabel` alone; now force thread reply mode only when an explicit thread target exists (`MessageThreadId`/`ReplyToId`), so DM `replyToModeByChatType.direct` overrides are honored outside real thread replies. (#26251) Thanks @dbachelder.
 - Slack/Threading: when `replyToMode="all"` auto-threads top-level Slack DMs, seed the thread session key from the message `ts` so the initial message and later replies share the same isolated `:thread:` session instead of falling back to base DM context. (#26849) Thanks @calder-sandy.
 - Agents/Subagents delivery: refactor subagent completion announce dispatch into an explicit queue/direct/fallback state machine, recover outbound channel-plugin resolution in cold/stale plugin-registry states across announce/message/gateway send paths, finalize cleanup bookkeeping when announce flow rejects, and treat Telegram sends without `message_id` as delivery failures (instead of false-success `"unknown"` IDs). (#26867, #25961, #26803, #25069, #26741) Thanks @SmithLabsLLC and @docaohieu2808.
-- Telegram/Webhook: pre-initialize webhook bots, switch webhook processing to callback-mode JSON handling, and preserve full near-limit payload reads under delayed handlers to prevent webhook request hangs and dropped updates. (#26156) Thanks @steipete.
+- Telegram/Webhook: pre-initialize webhook bots, switch webhook processing to callback-mode JSON handling, and preserve full near-limit payload reads under delayed handlers to prevent webhook request hangs and dropped updates. (#26156)
 - Slack/Session threads: prevent oversized parent-session inheritance from silently bricking new thread sessions, surface embedded context-overflow empty-result failures to users, and add configurable `session.parentForkMaxTokens` (default `100000`, `0` disables). (#26912) Thanks @markshields-tl.
 - Cron/Message multi-account routing: honor explicit `delivery.accountId` for isolated cron delivery resolution, and when `message.send` omits `accountId`, fall back to the sending agent's bound channel account instead of defaulting to the global account. (#27015, #26975) Thanks @lbo728 and @stakeswky.
 - Gateway/Message media roots: thread `agentId` through gateway `send` RPC and prefer explicit `agentId` over session/default resolution so non-default agent workspace media sends no longer fail with `LocalMediaAccessError`; added regression coverage for agent precedence and blank-agent fallback. (#23249) Thanks @Sid-Qin.
 - Followups/Routing: when explicit origin routing fails, allow same-channel fallback dispatch (while still blocking cross-channel fallback) so followup replies do not get dropped on transient origin-adapter failures. (#26109) Thanks @Sid-Qin.
-- Cron/Announce duplicate guard: track attempted announce/direct delivery separately from confirmed `delivered`, and suppress fallback main-session cron summaries when delivery was already attempted to avoid duplicate end-user sends in uncertain-ack paths. (#27018) Thanks @steipete.
+- Cron/Announce duplicate guard: track attempted announce/direct delivery separately from confirmed `delivered`, and suppress fallback main-session cron summaries when delivery was already attempted to avoid duplicate end-user sends in uncertain-ack paths. (#27018)
 - LINE/Lifecycle: keep LINE `startAccount` pending until abort so webhook startup is no longer misread as immediate channel exit, preventing restart-loop storms on LINE provider boot. (#26528) Thanks @Sid-Qin.
 - Discord/Gateway: capture and drain startup-time gateway `error` events before lifecycle listeners attach so early `Fatal Gateway error: 4014` closes surface as actionable intent guidance instead of uncaught gateway crashes. (#23832) Thanks @theotarr.
 - Discord/Inbound text: preserve embed `title` + `description` fallback text in message and forwarded snapshot parsing so embed titles are not silently dropped from agent input. (#26946) Thanks @stakeswky.
 - Slack/Inbound media fallback: deliver file-only messages even when Slack media downloads fail by adding a filename placeholder fallback, capping fallback names to the shared media-file limit, and normalizing empty filenames to `file` so attachment-only messages are not silently dropped. (#25181) Thanks @justinhuangcode.
-- Telegram/Preview cleanup: keep finalized text previews when a later assistant message is media-only (for example mixed text plus voice turns) by skipping finalized preview archival at assistant-message boundaries, preventing cleanup from deleting already-visible final text messages. (#27042) Thanks @steipete.
+- Telegram/Preview cleanup: keep finalized text previews when a later assistant message is media-only (for example mixed text plus voice turns) by skipping finalized preview archival at assistant-message boundaries, preventing cleanup from deleting already-visible final text messages. (#27042)
 - Telegram/Markdown spoilers: keep valid `||spoiler||` pairs while leaving unmatched trailing `||` delimiters as literal text, avoiding false all-or-nothing spoiler suppression. (#26105) Thanks @Sid-Qin.
 - Slack/Allowlist channels: match channel IDs case-insensitively during channel allowlist resolution so lowercase config keys (for example `c0abc12345`) correctly match Slack runtime IDs (`C0ABC12345`) under `groupPolicy: "allowlist"`, preventing silent channel-event drops. (#26878) Thanks @lbo728.
 - Discord/Typing indicator: prevent stuck typing indicators by sealing channel typing keepalive callbacks after idle/cleanup and ensuring Discord dispatch always marks typing idle even if preview-stream cleanup fails. (#26295) Thanks @ngutman.
 - Channels/Typing indicator: guard typing keepalive start callbacks after idle/cleanup close so post-close ticks cannot re-trigger stale typing indicators. (#26325) Thanks @win4r.
 - Followups/Typing indicator: ensure followup turns mark dispatch idle on every exit path (including `NO_REPLY`, empty payloads, and agent errors) so typing keepalive cleanup always runs and channel typing indicators do not get stuck after queued/silent followups. (#26881) Thanks @codexGW.
-- Voice-call/TTS tools: hide the `tts` tool when the message provider is `voice`, preventing voice-call runs from selecting self-playback TTS and falling into silent no-output loops. (#27025) Thanks @steipete.
-- Agents/Tools: normalize non-standard plugin tool results that omit `content` so embedded runs no longer crash with `Cannot read properties of undefined (reading 'filter')` after tool completion (including `tesseramemo_query`). (#27007) Thanks @steipete.
+- Voice-call/TTS tools: hide the `tts` tool when the message provider is `voice`, preventing voice-call runs from selecting self-playback TTS and falling into silent no-output loops. (#27025)
+- Agents/Tools: normalize non-standard plugin tool results that omit `content` so embedded runs no longer crash with `Cannot read properties of undefined (reading 'filter')` after tool completion (including `tesseramemo_query`). (#27007)
 - Agents/Tool-call dispatch: trim whitespace-padded tool names in both transcript repair and live streamed embedded-runner responses so exact-match tool lookup no longer fails with `Tool ... not found` for model outputs like `" read "`. (#27094) Thanks @openperf and @Sid-Qin.
 - Cron/Model overrides: when isolated `payload.model` is no longer allowlisted, fall back to default model selection instead of failing the job, while still returning explicit errors for invalid model strings. (#26717) Thanks @Youyou972.
 - Agents/Model fallback: keep explicit text + image fallback chains reachable even when `agents.defaults.models` allowlists are present, prefer explicit run `agentId` over session-key parsing for followup fallback override resolution (with session-key fallback), treat agent-level fallback overrides as configured in embedded runner preflight, and classify `model_cooldown` / `cooling down` errors as `rate_limit` so failover continues. (#11972, #24137, #17231)
@@ -3044,7 +3069,7 @@ Docs: https://docs.openclaw.ai
 
 ### Changes
 
-- Auto-reply/Abort shortcuts: expand standalone stop phrases (`stop openclaw`, `stop action`, `stop run`, `stop agent`, `please stop`, and related variants), accept trailing punctuation (for example `STOP OPENCLAW!!!`), add multilingual stop keywords (including ES/FR/ZH/HI/AR/JP/DE/PT/RU forms), and treat exact `do not do that` as a stop trigger while preserving strict standalone matching. (#25103) Thanks @steipete and @vincentkoc.
+- Auto-reply/Abort shortcuts: expand standalone stop phrases (`stop openclaw`, `stop action`, `stop run`, `stop agent`, `please stop`, and related variants), accept trailing punctuation (for example `STOP OPENCLAW!!!`), add multilingual stop keywords (including ES/FR/ZH/HI/AR/JP/DE/PT/RU forms), and treat exact `do not do that` as a stop trigger while preserving strict standalone matching. (#25103) Thanks @vincentkoc.
 - Android/App UX: ship a native four-step onboarding flow, move post-onboarding into a five-tab shell (Connect, Chat, Voice, Screen, Settings), add a full Connect setup/manual mode screen, and refresh Android chat/settings surfaces for the new navigation model.
 - Talk/Gateway config: add provider-agnostic Talk configuration with legacy compatibility, and expose gateway Talk ElevenLabs config metadata for setup/status surfaces.
 - Security/Audit: add `security.trust_model.multi_user_heuristic` to flag likely shared-user ingress and clarify the personal-assistant trust model, with hardening guidance for intentional multi-user setups (`sandbox.mode="all"`, workspace-scoped FS, reduced tool surface, no personal/private identities on shared runtimes).
@@ -3054,7 +3079,7 @@ Docs: https://docs.openclaw.ai
 
 - Routing/Session isolation: harden followup routing so explicit cross-channel origin replies never fall back to the active dispatcher on route failure, preserve queued overflow summary routing metadata (`channel`/`to`/`thread`) across followup drain, and prefer originating channel context over internal provider tags for embedded followup runs. This prevents webchat/control-ui context from hijacking Discord-targeted replies in shared sessions. (#25864) Thanks @Gamedesigner.
 - Security/Routing: fail closed for shared-session cross-channel replies by binding outbound target resolution to the current turn's source channel metadata (instead of stale session route fallbacks), and wire those turn-source fields through gateway + command delivery planners with regression coverage. (#24571) Thanks @brandonwise.
-- Heartbeat routing: prevent heartbeat leakage/spam into Discord and other direct-message destinations by blocking direct-chat heartbeat delivery targets and keeping blocked-delivery cron/exec prompts internal-only. (#25871) Thanks @steipete.
+- Heartbeat routing: prevent heartbeat leakage/spam into Discord and other direct-message destinations by blocking direct-chat heartbeat delivery targets and keeping blocked-delivery cron/exec prompts internal-only. (#25871)
 - Heartbeat defaults/prompts: switch the implicit heartbeat delivery target from `last` to `none` (opt-in for external delivery), and use internal-only cron/exec heartbeat prompt wording when delivery is disabled so background checks do not nudge user-facing relay behavior. (#25871, #24638, #25851)
 - Auto-reply/Heartbeat queueing: drop heartbeat runs when a session already has an active run instead of enqueueing a stale followup, preventing duplicate heartbeat response branches after queue drain. (#25610, #25606) Thanks @mcaxtr.
 - Cron/Heartbeat delivery: stop inheriting cached session `lastThreadId` for heartbeat-mode target resolution unless a thread/topic is explicitly requested, so announce-mode cron and heartbeat deliveries stay on top-level destinations instead of leaking into active conversation threads. (#25730) Thanks @markshields-tl.
@@ -3087,7 +3112,7 @@ Docs: https://docs.openclaw.ai
 - Windows/Media safety checks: align async local-file identity validation with sync-safe-open behavior by treating win32 `dev=0` stats as unknown-device fallbacks (while keeping strict dev checks when both sides are non-zero), fixing false `Local media path is not safe to read` drops for local attachments/TTS/images. (#25708, #21989, #25699, #25878) Thanks @kevinWangSheng.
 - iMessage/Reasoning safety: harden iMessage echo suppression with outbound `messageId` matching (plus scoped text fallback), and enforce reasoning-payload suppression on routed outbound delivery paths to prevent hidden thinking text from being sent as user-visible channel messages. (#25897, #1649, #25757) Thanks @rmarr and @Iranb.
 - Providers/OpenRouter/Auth profiles: bypass auth-profile cooldown/disable windows for OpenRouter, so provider failures no longer put OpenRouter profiles into local cooldown and stale legacy cooldown markers are ignored in fallback and status selection paths. (#25892) Thanks @alexanderatallah for raising this and @vincentkoc for the fix.
-- Providers/Google reasoning: sanitize invalid negative `thinkingBudget` payloads for Gemini 3.1 requests by dropping `-1` budgets and mapping configured reasoning effort to `thinkingLevel`, preventing malformed reasoning payloads on `google-generative-ai`. (#25900) Thanks @steipete.
+- Providers/Google reasoning: sanitize invalid negative `thinkingBudget` payloads for Gemini 3.1 requests by dropping `-1` budgets and mapping configured reasoning effort to `thinkingLevel`, preventing malformed reasoning payloads on `google-generative-ai`. (#25900)
 - Providers/SiliconFlow: normalize `thinking="off"` to `thinking: null` for `Pro/*` model payloads to avoid provider-side 400 loops and misleading compaction retries. (#25435) Thanks @Zjianru.
 - Models/Bedrock auth: normalize additional Bedrock provider aliases (`bedrock`, `aws-bedrock`, `aws_bedrock`, `amazon bedrock`) to canonical `amazon-bedrock`, ensuring auth-mode resolution consistently selects AWS SDK fallback. (#25756) Thanks @fwhite13.
 - Models/Providers: preserve explicit user `reasoning` overrides when merging provider model config with built-in catalog metadata, so `reasoning: false` is no longer overwritten by catalog defaults. (#25314) Thanks @lbo728.
@@ -3183,7 +3208,7 @@ Docs: https://docs.openclaw.ai
 - Providers/Groq: avoid classifying Groq TPM limit errors as context overflow so throttling paths no longer trigger overflow recovery logic. (#16176) Thanks @dddabtc.
 - Gateway/Restart: treat child listener PIDs as owned by the service runtime PID during restart health checks to avoid false stale-process kills and restart timeouts on launchd/systemd. (#24696) Thanks @gumadeiras.
 - Config/Write: apply `unsetPaths` with immutable path-copy updates so config writes never mutate caller-provided objects, and harden `openclaw config get/set/unset` path traversal by rejecting prototype-key segments and inherited-property traversal. (#24134) thanks @frankekn.
-- Channels/WhatsApp: accept `channels.whatsapp.enabled` in config validation to match built-in channel auto-enable behavior, preventing `Unrecognized key: "enabled"` failures during channel setup. (#24263) Thanks @steipete.
+- Channels/WhatsApp: accept `channels.whatsapp.enabled` in config validation to match built-in channel auto-enable behavior, preventing `Unrecognized key: "enabled"` failures during channel setup. (#24263)
 - Security/Exec: detect obfuscated commands before exec allowlist decisions and require explicit approval for obfuscation patterns. (#8592) Thanks @CornBrother0x and @vincentkoc.
 - Security/ACP: harden ACP client permission auto-approval to require trusted core tool IDs, ignore untrusted `toolCall.kind` hints, and scope `read` auto-approval to the active working directory so unknown tool names and out-of-scope file reads always prompt. Thanks @nedlir for reporting.
 - Security/Skills: escape user-controlled prompt, filename, and output-path values in `openai-image-gen` HTML gallery generation to prevent stored XSS in generated `index.html` output. (#12538) Thanks @CornBrother0x.
@@ -3208,7 +3233,7 @@ Docs: https://docs.openclaw.ai
 - Update/Core: add an optional built-in auto-updater for package installs (`update.auto.*`), default-off, with stable rollout delay+jitter and beta hourly cadence.
 - CLI/Update: add `openclaw update --dry-run` to preview channel/tag/target/restart actions without mutating config, installing, syncing plugins, or restarting.
 - Config/UI: add tag-aware settings filtering and broaden config labels/help copy so fields are easier to discover and understand in the dashboard config screen.
-- Channels/Synology Chat: add a native Synology Chat channel plugin with webhook ingress, direct-message routing, outbound send/media support, per-account config, and DM policy controls. (#23012) Thanks @steipete.
+- Channels/Synology Chat: add a native Synology Chat channel plugin with webhook ingress, direct-message routing, outbound send/media support, per-account config, and DM policy controls. (#23012)
 - iOS/Talk: prefetch TTS segments and suppress expected speech-cancellation errors for smoother talk playback. (#22833) Thanks @ngutman.
 - Memory/FTS: add Spanish and Portuguese stop-word filtering for query expansion in FTS-only search mode, improving conversational recall for both languages. Thanks @vincentkoc.
 - Memory/FTS: add Japanese-aware query expansion tokenization and stop-word filtering (including mixed-script terms like ASCII + katakana) for FTS-only search mode. Thanks @vincentkoc.
@@ -3230,10 +3255,10 @@ Docs: https://docs.openclaw.ai
 - Agents/Moonshot: force `supportsDeveloperRole=false` for Moonshot-compatible `openai-completions` models (provider `moonshot` and Moonshot base URLs), so initial runs no longer send unsupported `developer` roles that trigger `ROLE_UNSPECIFIED` errors. (#21060, #22194) Thanks @ShengFuC.
 - Agents/Kimi: classify Moonshot `Your request exceeded model token limit` failures as context overflows so auto-compaction and user-facing overflow recovery trigger correctly instead of surfacing raw invalid-request errors. (#9562) Thanks @danilofalcao.
 - Providers/Moonshot: mark Kimi K2.5 as image-capable in implicit + onboarding model definitions, and refresh stale explicit provider capability fields (`input`/`reasoning`/context limits) from implicit catalogs so existing configs pick up Moonshot vision support without manual model rewrites. (#13135, #4459) Thanks @manikv12.
-- Agents/Transcript: enable consecutive-user turn merging for strict non-OpenAI `openai-completions` providers (for example Moonshot/Kimi), reducing `roles must alternate` ordering failures on OpenAI-compatible endpoints while preserving current OpenRouter/Opencode behavior. (#7693) Thanks @steipete.
+- Agents/Transcript: enable consecutive-user turn merging for strict non-OpenAI `openai-completions` providers (for example Moonshot/Kimi), reducing `roles must alternate` ordering failures on OpenAI-compatible endpoints while preserving current OpenRouter/Opencode behavior. (#7693)
 - Install/Discord Voice: make the native Opus decoder optional so `openclaw` install/update no longer hard-fails when native builds fail, while keeping `opusscript` as the runtime fallback decoder for Discord voice flows. (#23737, #23733, #23703) Thanks @jeadland, @Sheetaa, and @Breakyman.
 - Docker/Setup: precreate `$OPENCLAW_CONFIG_DIR/identity` during `docker-setup.sh` so CLI commands that need device identity (for example `devices list`) avoid `EACCES ... /home/node/.openclaw/identity` failures on restrictive bind mounts. (#23948) Thanks @ackson-beep.
-- Exec/Background: stop applying the default exec timeout to background sessions (`background: true` or explicit `yieldMs`) when no explicit timeout is set, so long-running background jobs are no longer terminated at the default timeout boundary. (#23303) Thanks @steipete.
+- Exec/Background: stop applying the default exec timeout to background sessions (`background: true` or explicit `yieldMs`) when no explicit timeout is set, so long-running background jobs are no longer terminated at the default timeout boundary. (#23303)
 - Slack/Threading: sessions: keep parent-session forking and thread-history context active beyond first turn by removing first-turn-only gates in session init, thread-history fetch, and reply prompt context injection. (#23843, #23090) Thanks @vincentkoc and @Taskle.
 - Slack/Threading: respect `replyToMode` when Slack auto-populates top-level `thread_ts`, and ignore inline `replyToId` directive tags when `replyToMode` is `off` so thread forcing stays disabled unless explicitly configured. (#23839, #23320, #23513) Thanks @vincentkoc and @dorukardahan.
 - Slack/Extension: forward `message read` `threadId` to `readMessages` and use delivery-context `threadId` as outbound `thread_ts` fallback so extension replies/reads stay in the correct Slack thread. (#22216, #22485, #23836) Thanks @vincentkoc, @lan17 and @dorukardahan.
@@ -3253,7 +3278,7 @@ Docs: https://docs.openclaw.ai
 - Telegram/Webhook: add `channels.telegram.webhookPort` config support and pass it through plugin startup wiring to the monitor listener.
 - Browser/Extension Relay: refactor the MV3 worker to preserve debugger attachments across relay drops, auto-reconnect with bounded backoff+jitter, persist and rehydrate attached tab state via `chrome.storage.session`, recover from `target_closed` navigation detaches, guard stale socket handlers, enforce per-tab operation locks and per-request timeouts, and add lifecycle keepalive/badge refresh hooks (`alarms`, `webNavigation`). (#15099, #6175, #8468, #9807)
 - Browser/Relay: treat extension websocket as connected only when `OPEN`, allow reconnect when a stale `CLOSING/CLOSED` extension socket lingers, and guard stale socket message/close handlers so late events cannot clear active relay state; includes regression coverage for live-duplicate `409` rejection and immediate reconnect-after-close races. (#15099, #18698, #20688)
-- Browser/Remote CDP: extend stale-target recovery so `ensureTabAvailable()` now reuses the sole available tab for remote CDP profiles (same behavior as extension profiles) while preserving strict `tab not found` errors when multiple tabs exist; includes remote-profile regression tests. (#15989) Thanks @steipete.
+- Browser/Remote CDP: extend stale-target recovery so `ensureTabAvailable()` now reuses the sole available tab for remote CDP profiles (same behavior as extension profiles) while preserving strict `tab not found` errors when multiple tabs exist; includes remote-profile regression tests. (#15989)
 - Gateway/Pairing: treat `operator.admin` as satisfying other `operator.*` scope checks during device-auth verification so local CLI/TUI sessions stop entering pairing-required loops for pairing/approval-scoped commands. (#22062, #22193, #21191) Thanks @Botaccess, @jhartshorn, and @ctbritt.
 - Gateway/Pairing: auto-approve loopback `scope-upgrade` pairing requests (including device-token reconnects) so local clients do not disconnect on pairing-required scope elevation. (#23708) Thanks @widingmarcus-cyber.
 - Gateway/Scopes: include `operator.read` and `operator.write` in default operator connect scope bundles across CLI, Control UI, and macOS clients so write-scoped announce/sub-agent follow-up calls no longer hit `pairing required` disconnects on loopback gateways. (#22582) thanks @YuzuruS.
@@ -3291,25 +3316,25 @@ Docs: https://docs.openclaw.ai
 - Security/Group policy: harden `channels.*.groups.*.toolsBySender` matching by requiring explicit sender-key types (`id:`, `e164:`, `username:`, `name:`), preventing cross-identifier collisions across mutable/display-name fields while keeping legacy untyped keys on a deprecated ID-only path. Thanks @jiseoung for reporting.
 - Channels/Group policy: fail closed when `groupPolicy: "allowlist"` is set without explicit `groups`, honor account-level `groupPolicy` overrides, and enforce `groupPolicy: "disabled"` as a hard group block. (#22215) Thanks @etereo.
 - Telegram/Discord extensions: propagate trusted `mediaLocalRoots` through extension outbound `sendMedia` options so extension direct-send media paths honor agent-scoped local-media allowlists. (#20029, #21903, #23227)
-- Agents/Exec: honor explicit agent context when resolving `tools.exec` defaults for runs with opaque/non-agent session keys, so per-agent `host/security/ask` policies are applied consistently. (#11832) Thanks @steipete.
+- Agents/Exec: honor explicit agent context when resolving `tools.exec` defaults for runs with opaque/non-agent session keys, so per-agent `host/security/ask` policies are applied consistently. (#11832)
 - CLI/Sessions: resolve implicit session-store path templates with the configured default agent ID so named-agent setups do not silently read/write stale `agent:main` session/auth stores. (#22685) Thanks @sene1337.
-- Doctor/Security: add an explicit warning that `approvals.exec.enabled=false` disables forwarding only, while enforcement remains driven by host-local `exec-approvals.json` policy. (#15047) Thanks @steipete.
-- Sandbox/Docker: default sandbox container user to the workspace owner `uid:gid` when `agents.*.sandbox.docker.user` is unset, fixing non-root gateway file-tool permissions under capability-dropped containers. (#20979) Thanks @steipete.
+- Doctor/Security: add an explicit warning that `approvals.exec.enabled=false` disables forwarding only, while enforcement remains driven by host-local `exec-approvals.json` policy. (#15047)
+- Sandbox/Docker: default sandbox container user to the workspace owner `uid:gid` when `agents.*.sandbox.docker.user` is unset, fixing non-root gateway file-tool permissions under capability-dropped containers. (#20979)
 - Plugins/Media sandbox: propagate trusted `mediaLocalRoots` through plugin action dispatch (including Discord/Telegram action adapters) so plugin send paths enforce the same agent-scoped local-media sandbox roots as core outbound sends. (#20258, #22718)
-- Agents/Workspace guard: map sandbox container-workdir file-tool paths (for example `/workspace/...` and `file:///workspace/...`) to host workspace roots before workspace-only validation, preventing false `Path escapes sandbox root` rejections for sandbox file tools. (#9560) Thanks @steipete.
-- Gateway/Exec approvals: expire approval requests immediately when no approval-capable gateway clients are connected and no forwarding targets are available, avoiding delayed approvals after restarts/offline approver windows. (#22144) Thanks @steipete.
+- Agents/Workspace guard: map sandbox container-workdir file-tool paths (for example `/workspace/...` and `file:///workspace/...`) to host workspace roots before workspace-only validation, preventing false `Path escapes sandbox root` rejections for sandbox file tools. (#9560)
+- Gateway/Exec approvals: expire approval requests immediately when no approval-capable gateway clients are connected and no forwarding targets are available, avoiding delayed approvals after restarts/offline approver windows. (#22144)
 - Security/Exec approvals: when approving wrapper commands with allow-always in allowlist mode, persist inner executable paths for known dispatch wrappers (`env`, `nice`, `nohup`, `stdbuf`, `timeout`) and fail closed (no persisted entry) when wrapper unwrapping is not safe, preventing wrapper-path approval bypasses. Thanks @tdjackey for reporting.
-- Node/macOS exec host: default headless macOS node `system.run` to local execution and only route through the companion app when `OPENCLAW_NODE_EXEC_HOST=app` is explicitly set, avoiding companion-app filesystem namespace mismatches during exec. (#23547) Thanks @steipete.
+- Node/macOS exec host: default headless macOS node `system.run` to local execution and only route through the companion app when `OPENCLAW_NODE_EXEC_HOST=app` is explicitly set, avoiding companion-app filesystem namespace mismatches during exec. (#23547)
 - Sandbox/Media: map container workspace paths (`/workspace/...` and `file:///workspace/...`) back to the host sandbox root for outbound media validation, preventing false deny errors for sandbox-generated local media. (#23083) Thanks @echo931.
 - Sandbox/Docker: apply custom bind mounts after workspace mounts and prioritize bind-source resolution on overlapping paths, so explicit workspace binds are no longer ignored. (#22669) Thanks @tasaankaeris.
 - Exec approvals/Forwarding: restore Discord text forwarding when component approvals are not configured, and carry request snapshots through resolve events so resolved notices still forward after cache misses/restarts. (#22988) Thanks @bubmiller.
 - Control UI/WebSocket: stop and clear the browser gateway client on UI teardown so remounts cannot leave orphan websocket clients that create duplicate active connections. (#23422) Thanks @floatinggball-design.
 - Control UI/WebSocket: send a stable per-tab `instanceId` in websocket connect frames so reconnect cycles keep a consistent client identity for diagnostics and presence tracking. (#23616) Thanks @zq58855371-ui.
 - Config/Memory: allow `"mistral"` in `agents.defaults.memorySearch.provider` and `agents.defaults.memorySearch.fallback` schema validation. (#14934) Thanks @ThomsenDrake.
-- Feishu/Commands: in group chats, command authorization now falls back to top-level `channels.feishu.allowFrom` when per-group `allowFrom` is not set, so `/command` no longer gets blocked by an unintended empty allowlist. (#23756) Thanks @steipete.
+- Feishu/Commands: in group chats, command authorization now falls back to top-level `channels.feishu.allowFrom` when per-group `allowFrom` is not set, so `/command` no longer gets blocked by an unintended empty allowlist. (#23756)
 - Dev tooling: prevent `CLAUDE.md` symlink target regressions by excluding CLAUDE symlink sentinels from `oxfmt` and marking them `-text` in `.gitattributes`, so formatter/EOL normalization cannot reintroduce trailing-newline targets. Thanks @vincentkoc.
 - Agents/Compaction: restore embedded compaction safeguard/context-pruning extension loading in production by wiring bundled extension factories into the resource loader instead of runtime file-path resolution. (#22349; landed from contributor PR #5005 by @Diaspar4u) Thanks @Diaspar4u.
-- Feishu/Media: for inbound video messages that include both `file_key` (video) and `image_key` (thumbnail), prefer `file_key` when downloading media so video attachments are saved instead of silently failing on thumbnail keys. (#23633) Thanks @steipete.
+- Feishu/Media: for inbound video messages that include both `file_key` (video) and `image_key` (thumbnail), prefer `file_key` when downloading media so video attachments are saved instead of silently failing on thumbnail keys. (#23633)
 - Hooks/Loader: avoid redundant hook-module recompilation on gateway restart by skipping cache-busting for bundled hooks and using stable file metadata keys (`mtime+size`) for mutable workspace/managed/plugin hook imports. (#16953) Thanks @mudrii.
 - Hooks/Cron: suppress duplicate main-session events for delivered hook turns and mark `SILENT_REPLY_TOKEN` (`NO_REPLY`) early exits as delivered to prevent hook context pollution. (#20678) Thanks @JonathanWorks.
 - Providers/OpenRouter: inject `cache_control` on system prompts for OpenRouter Anthropic models to improve prompt-cache reuse. (#17473) Thanks @rrenamed.
@@ -3788,7 +3813,7 @@ Docs: https://docs.openclaw.ai
 - Ollama/Qwen: handle Qwen 3 reasoning field format in Ollama responses. (#18631) Thanks @mr-sk.
 - OpenAI/Transcripts: always drop orphaned reasoning blocks from transcript repair. (#18632) Thanks @TySabs.
 - Fix types in all tests. Typecheck the whole repository.
-- Gateway/Channels: wire `gateway.channelHealthCheckMinutes` into strict config validation, treat implicit account status as managed for health checks, and harden channel auto-restart flow (preserve restart-attempt caps across crash loops, propagate enabled/configured runtime flags, and stop pending restart backoff after manual stop). Thanks @steipete.
+- Gateway/Channels: wire `gateway.channelHealthCheckMinutes` into strict config validation, treat implicit account status as managed for health checks, and harden channel auto-restart flow (preserve restart-attempt caps across crash loops, propagate enabled/configured runtime flags, and stop pending restart backoff after manual stop).
 - Gateway/WebChat: hard-cap `chat.history` oversized payloads by truncating high-cost fields and replacing over-budget entries with placeholders, so history fetches stay within configured byte limits and avoid chat UI freezes. (#18505)
 - UI/Usage: replace lingering undefined `var(--text-muted)` usage with `var(--muted)` in usage date-range and chart styles to keep muted text visible across themes. (#17975) Thanks @jogelin.
 - UI/Usage: preserve selected-range totals when timeline data is downsampled by bucket-aggregating timeseries points (instead of dropping intermediate points), so filtered tokens/cost stay accurate. (#17959) Thanks @jogelin.
@@ -4798,21 +4823,21 @@ Docs: https://docs.openclaw.ai
 
 - Providers: Ollama discovery + docs; Venice guide upgrades + cross-links. (#1606) Thanks @abhaymundhara. https://docs.openclaw.ai/providers/ollama https://docs.openclaw.ai/providers/venice
 - Channels: LINE plugin (Messaging API) with rich replies + quick replies. (#1630) Thanks @plum-dawg.
-- TTS: Edge fallback (keyless) + `/tts` auto modes. (#1668, #1667) Thanks @steipete, @sebslight. https://docs.openclaw.ai/tts
+- TTS: Edge fallback (keyless) + `/tts` auto modes. (#1668, #1667) Thanks @sebslight. https://docs.openclaw.ai/tts
 - Exec approvals: approve in-chat via `/approve` across all channels (including plugins). (#1621) Thanks @czekaj. https://docs.openclaw.ai/tools/exec-approvals https://docs.openclaw.ai/tools/slash-commands
 - Telegram: DM topics as separate sessions + outbound link preview toggle. (#1597, #1700) Thanks @rohannagpal, @zerone0x. https://docs.openclaw.ai/channels/telegram
 
 ### Changes
 
 - Channels: add LINE plugin (Messaging API) with rich replies, quick replies, and plugin HTTP registry. (#1630) Thanks @plum-dawg.
-- TTS: add Edge TTS provider fallback, defaulting to keyless Edge with MP3 retry on format failures. (#1668) Thanks @steipete. https://docs.openclaw.ai/tts
+- TTS: add Edge TTS provider fallback, defaulting to keyless Edge with MP3 retry on format failures. (#1668) https://docs.openclaw.ai/tts
 - TTS: add auto mode enum (off/always/inbound/tagged) with per-session `/tts` override. (#1667) Thanks @sebslight. https://docs.openclaw.ai/tts
 - Telegram: treat DM topics as separate sessions and keep DM history limits stable with thread suffixes. (#1597) Thanks @rohannagpal.
 - Telegram: add `channels.telegram.linkPreview` to toggle outbound link previews. (#1700) Thanks @zerone0x. https://docs.openclaw.ai/channels/telegram
 - Web search: add Brave freshness filter parameter for time-scoped results. (#1688) Thanks @JonUleis. https://docs.openclaw.ai/tools/web
 - UI: refresh Control UI dashboard design system (colors, icons, typography). (#1745, #1786) Thanks @EnzeD, @mousberg.
 - Exec approvals: forward approval prompts to chat with `/approve` for all channels (including plugins). (#1621) Thanks @czekaj. https://docs.openclaw.ai/tools/exec-approvals https://docs.openclaw.ai/tools/slash-commands
-- Gateway: expose config.patch in the gateway tool with safe partial updates + restart sentinel. (#1653) Thanks @steipete.
+- Gateway: expose config.patch in the gateway tool with safe partial updates + restart sentinel. (#1653)
 - Diagnostics: add diagnostic flags for targeted debug logs (config + env override). https://docs.openclaw.ai/diagnostics/flags
 - Docs: expand FAQ (migration, scheduling, concurrency, model recommendations, OpenAI subscription auth, Pi sizing, hackable install, docs SSL workaround).
 - Docs: add verbose installer troubleshooting guidance.
@@ -4825,9 +4850,9 @@ Docs: https://docs.openclaw.ai
 
 - Web UI: fix config/debug layout overflow, scrolling, and code block sizing. (#1715) Thanks @saipreetham589.
 - Web UI: show Stop button during active runs, swap back to New session when idle. (#1664) Thanks @ndbroadbent.
-- Web UI: clear stale disconnect banners on reconnect; allow form saves with unsupported schema paths but block missing schema. (#1707) Thanks @steipete.
+- Web UI: clear stale disconnect banners on reconnect; allow form saves with unsupported schema paths but block missing schema. (#1707)
 - Web UI: hide internal `message_id` hints in chat bubbles.
-- Gateway: allow Control UI token-only auth to skip device pairing even when device identity is present (`gateway.controlUi.allowInsecureAuth`). (#1679) Thanks @steipete.
+- Gateway: allow Control UI token-only auth to skip device pairing even when device identity is present (`gateway.controlUi.allowInsecureAuth`). (#1679)
 - Matrix: decrypt E2EE media attachments with preflight size guard. (#1744) Thanks @araa47.
 - BlueBubbles: route phone-number targets to DMs, avoid leaking routing IDs, and auto-create missing DMs (Private API required). (#1751) Thanks @tyler6204. https://docs.openclaw.ai/channels/bluebubbles
 - BlueBubbles: keep part-index GUIDs in reply tags when short IDs are missing.
@@ -4900,7 +4925,7 @@ Docs: https://docs.openclaw.ai
 - Heartbeat: accept plugin channel ids for heartbeat target validation + UI hints.
 - Messaging/Sessions: mirror outbound sends into target session keys (threads + dmScope), create session entries on send, and normalize session key casing. (#1520, commit 4b6cdd1d3)
 - Sessions: reject array-backed session stores to prevent silent wipes. (#1469)
-- Gateway: compare Linux process start time to avoid PID recycling lock loops; keep locks unless stale. (#1572) Thanks @steipete.
+- Gateway: compare Linux process start time to avoid PID recycling lock loops; keep locks unless stale. (#1572)
 - Gateway: accept null optional fields in exec approval requests. (#1511) Thanks @pvoo.
 - Exec approvals: persist allowlist entry ids to keep macOS allowlist rows stable. (#1521) Thanks @ngutman.
 - Exec: honor tools.exec ask/security defaults for elevated approvals (avoid unwanted prompts). (commit 5662a9cdf)
@@ -5206,7 +5231,7 @@ Docs: https://docs.openclaw.ai
 - macOS: bundle Textual resources in packaged app builds to avoid code block crashes. (#1006)
 - Daemon: include HOME in service environments to avoid missing HOME errors. (#1214)
 
-Thanks @AlexMikhalev, @CoreyH, @John-Rood, @KrauseFx, @MaudeBot, @Nachx639, @NicholaiVogel, @RyanLisse, @ThePickle31, @VACInc, @Whoaa512, @YuriNachos, @aaronveklabs, @abdaraxus, @alauppe, @ameno-, @artuskg, @austinm911, @bradleypriest, @cheeeee, @dougvk, @fogboots, @gnarco, @gumadeiras, @jdrhyne, @joelklabo, @longmaba, @mukhtharcm, @odysseus0, @oscargavin, @rhjoh, @sebslight, @sibbl, @sleontenko, @steipete, @suminhthanh, @thewilloftheshadow, @tyler6204, @vignesh07, @visionik, @ysqander, @zerone0x.
+Thanks @AlexMikhalev, @CoreyH, @John-Rood, @KrauseFx, @MaudeBot, @Nachx639, @NicholaiVogel, @RyanLisse, @ThePickle31, @VACInc, @Whoaa512, @YuriNachos, @aaronveklabs, @abdaraxus, @alauppe, @ameno-, @artuskg, @austinm911, @bradleypriest, @cheeeee, @dougvk, @fogboots, @gnarco, @gumadeiras, @jdrhyne, @joelklabo, @longmaba, @mukhtharcm, @odysseus0, @oscargavin, @rhjoh, @sebslight, @sibbl, @sleontenko, @suminhthanh, @thewilloftheshadow, @tyler6204, @vignesh07, @visionik, @ysqander, @zerone0x.
 
 ### Breaking
 
@@ -5575,7 +5600,7 @@ Thanks @AlexMikhalev, @CoreyH, @John-Rood, @KrauseFx, @MaudeBot, @Nachx639, @Nic
 - Gateway/CLI: honor `CLAWDBOT_LAUNCHD_LABEL` / `CLAWDBOT_SYSTEMD_UNIT` overrides; `agents.list` respects explicit config; reduce noisy loopback WS logs during tests; run `openclaw doctor --non-interactive` during updates. (#781) - thanks @ronyrus.
 - Onboarding/Control UI: refuse invalid configs (run doctor first); quote Windows browser URLs for OAuth; keep chat scroll position unless the user is near the bottom. (#764) - thanks @mukhtharcm; (#794) - thanks @roshanasingh4; (#217) - thanks @thewilloftheshadow.
 - Tools/UI: harden tool input schemas for strict providers; drop null-only union variants for Gemini schema cleanup; treat `maxChars: 0` as unlimited; keep TUI last streamed response instead of "(no output)". (#782) - thanks @AbhisekBasu1; (#796) - thanks @gabriel-trigo; (#747) - thanks @thewilloftheshadow.
-- Connections UI: polish multi-account account cards. (#816) - thanks @steipete.
+- Connections UI: polish multi-account account cards. (#816)
 
 ### Installer
 
@@ -5615,7 +5640,7 @@ Thanks @AlexMikhalev, @CoreyH, @John-Rood, @KrauseFx, @MaudeBot, @Nachx639, @Nic
 - Tests: add Docker plugin loader + tgz-install smoke test.
 - Tests: extend Docker plugin E2E to cover installing from local folders (`plugins.load.paths`) and `file:` npm specs.
 - Tests: add coverage for pre-compaction memory flush settings.
-- Tests: modernize live model smoke selection for current releases and enforce tools/images/thinking-high coverage. (#769) - thanks @steipete.
+- Tests: modernize live model smoke selection for current releases and enforce tools/images/thinking-high coverage. (#769)
 - Agents/Tools: add `apply_patch` tool for multi-file edits (experimental; gated by tools.exec.applyPatch; OpenAI-only).
 - Agents/Tools: rename the bash tool to exec (config alias maintained). (#748) - thanks @myfunc.
 - Agents: add pre-compaction memory flush config (`agents.defaults.compaction.*`) with a soft threshold + system prompt.
@@ -5635,8 +5660,8 @@ Thanks @AlexMikhalev, @CoreyH, @John-Rood, @KrauseFx, @MaudeBot, @Nachx639, @Nic
 ### Fixes
 
 - Models/Onboarding: configure MiniMax (minimax.io) via Anthropic-compatible `/anthropic` endpoint by default (keep `minimax-api` as a legacy alias).
-- Models: normalize Gemini 3 Pro/Flash IDs to preview names for live model lookups. (#769) - thanks @steipete.
-- CLI: fix guardCancel typing for configure prompts. (#769) - thanks @steipete.
+- Models: normalize Gemini 3 Pro/Flash IDs to preview names for live model lookups. (#769)
+- CLI: fix guardCancel typing for configure prompts. (#769)
 - Gateway/WebChat: include handshake validation details in the WebSocket close reason for easier debugging; preserve close codes.
 - Gateway/Auth: send invalid connect responses before closing the handshake; stabilize invalid-connect auth test.
 - Gateway: tighten gateway listener detection.
@@ -5653,7 +5678,7 @@ Thanks @AlexMikhalev, @CoreyH, @John-Rood, @KrauseFx, @MaudeBot, @Nachx639, @Nic
 - Auto-reply: align `/think` default display with model reasoning defaults. (#751) - thanks @gabriel-trigo.
 - Auto-reply: flush block reply buffers on tool boundaries. (#750) - thanks @sebslight.
 - Auto-reply: allow sender fallback for command authorization when `SenderId` is empty (WhatsApp self-chat). (#755) - thanks @juanpablodlc.
-- Auto-reply: treat whitespace-only sender ids as missing for command authorization (WhatsApp self-chat). (#766) - thanks @steipete.
+- Auto-reply: treat whitespace-only sender ids as missing for command authorization (WhatsApp self-chat). (#766)
 - Heartbeat: refresh prompt text for updated defaults.
 - Memory/QMD: prefer `qmd collection add --glob` for current QMD releases and fall back to legacy `--mask` when older builds reject it. (#55123) Thanks @ForceConstant and @vincentkoc.
 - Agents/Tools: use PowerShell on Windows to capture system utility output. (#748) - thanks @myfunc.

INCIDENT_RESPONSE.md

@@ -0,0 +1,53 @@
+# OpenClaw Incident Response Plan
+
+## 1. Detection and triage
+
+We monitor security signals from:
+
+- GitHub Security Advisories (GHSA) and private vulnerability reports.
+- Public GitHub issues/discussions when reports are not sensitive.
+- Official plublic discussion groups and channels (i.e. Discord and X).
+- Automated signals (for example Dependabot, CodeQL, npm advisories, and secret scanning).
+
+Initial triage:
+
+1. Confirm affected component, version, and trust boundary impact.
+2. Classify as security issue vs hardening/no-action using the repository `SECURITY.md` scope and out-of-scope rules.
+3. An incident owner responds accordingly.
+
+## 2. Assessment
+
+Severity guide:
+
+- **Critical:** Package/release/repository compromise, active exploitation, or unauthenticated trust-boundary bypass with high-impact control or data exposure.
+- **High:** Verified trust-boundary bypass requiring limited preconditions (for example authenticated but unauthorized high-impact action), or exposure of OpenClaw-owned sensitive credentials.
+- **Medium:** Significant security weakness with practical impact but constrained exploitability or substantial prerequisites.
+- **Low:** Defense-in-depth findings, narrowly scoped denial-of-service, or hardening/parity gaps without a demonstrated trust-boundary bypass.
+
+## 3. Response
+
+1. Acknowledge receipt to the reporter (private when sensitive).
+2. Reproduce on supported releases and latest `main`, then implement and validate a patch with regression coverage.
+3. For critical/high incidents, prepare patched release(s) as fast as practical.
+4. For medium/low incidents, patch in normal release flow and document mitigation guidance.
+
+## 4. Communication
+
+We communicate through:
+
+- GitHub Security Advisories in the affected repository.
+- Release notes/changelog entries for fixed versions.
+- Direct reporter follow-up on status and resolution.
+
+Disclosure policy:
+
+- Critical/high incidents should receive coordinated disclosure, with CVE issuance when appropriate.
+- Low-risk hardening findings may be documented in release notes or advisories without CVE, depending on impact and user exposure.
+
+## 5. Recovery and follow-up
+
+After shipping the fix:
+
+1. Verify remediations in CI and release artifacts.
+2. Run a short post-incident review (timeline, root cause, detection gap, prevention plan).
+3. Add follow-up hardening/tests/docs tasks and track them to completion.

apps/macos/Sources/OpenClawProtocol/GatewayModels.swift

@@ -1893,6 +1893,7 @@ public struct ConfigApplyParams: Codable, Sendable {
     public let raw: String
     public let basehash: String?
     public let sessionkey: String?
+    public let deliverycontext: [String: AnyCodable]?
     public let note: String?
     public let restartdelayms: Int?
 
@@ -1900,12 +1901,14 @@ public struct ConfigApplyParams: Codable, Sendable {
         raw: String,
         basehash: String?,
         sessionkey: String?,
+        deliverycontext: [String: AnyCodable]?,
         note: String?,
         restartdelayms: Int?)
     {
         self.raw = raw
         self.basehash = basehash
         self.sessionkey = sessionkey
+        self.deliverycontext = deliverycontext
         self.note = note
         self.restartdelayms = restartdelayms
     }
@@ -1914,6 +1917,7 @@ public struct ConfigApplyParams: Codable, Sendable {
         case raw
         case basehash = "baseHash"
         case sessionkey = "sessionKey"
+        case deliverycontext = "deliveryContext"
         case note
         case restartdelayms = "restartDelayMs"
     }
@@ -1923,6 +1927,7 @@ public struct ConfigPatchParams: Codable, Sendable {
     public let raw: String
     public let basehash: String?
     public let sessionkey: String?
+    public let deliverycontext: [String: AnyCodable]?
     public let note: String?
     public let restartdelayms: Int?
 
@@ -1930,12 +1935,14 @@ public struct ConfigPatchParams: Codable, Sendable {
         raw: String,
         basehash: String?,
         sessionkey: String?,
+        deliverycontext: [String: AnyCodable]?,
         note: String?,
         restartdelayms: Int?)
     {
         self.raw = raw
         self.basehash = basehash
         self.sessionkey = sessionkey
+        self.deliverycontext = deliverycontext
         self.note = note
         self.restartdelayms = restartdelayms
     }
@@ -1944,6 +1951,7 @@ public struct ConfigPatchParams: Codable, Sendable {
         case raw
         case basehash = "baseHash"
         case sessionkey = "sessionKey"
+        case deliverycontext = "deliveryContext"
         case note
         case restartdelayms = "restartDelayMs"
     }
@@ -4313,17 +4321,20 @@ public struct ChatEvent: Codable, Sendable {
 
 public struct UpdateRunParams: Codable, Sendable {
     public let sessionkey: String?
+    public let deliverycontext: [String: AnyCodable]?
     public let note: String?
     public let restartdelayms: Int?
     public let timeoutms: Int?
 
     public init(
         sessionkey: String?,
+        deliverycontext: [String: AnyCodable]?,
         note: String?,
         restartdelayms: Int?,
         timeoutms: Int?)
     {
         self.sessionkey = sessionkey
+        self.deliverycontext = deliverycontext
         self.note = note
         self.restartdelayms = restartdelayms
         self.timeoutms = timeoutms
@@ -4331,6 +4342,7 @@ public struct UpdateRunParams: Codable, Sendable {
 
     private enum CodingKeys: String, CodingKey {
         case sessionkey = "sessionKey"
+        case deliverycontext = "deliveryContext"
         case note
         case restartdelayms = "restartDelayMs"
         case timeoutms = "timeoutMs"

apps/shared/OpenClawKit/Sources/OpenClawProtocol/GatewayModels.swift

@@ -1893,6 +1893,7 @@ public struct ConfigApplyParams: Codable, Sendable {
     public let raw: String
     public let basehash: String?
     public let sessionkey: String?
+    public let deliverycontext: [String: AnyCodable]?
     public let note: String?
     public let restartdelayms: Int?
 
@@ -1900,12 +1901,14 @@ public struct ConfigApplyParams: Codable, Sendable {
         raw: String,
         basehash: String?,
         sessionkey: String?,
+        deliverycontext: [String: AnyCodable]?,
         note: String?,
         restartdelayms: Int?)
     {
         self.raw = raw
         self.basehash = basehash
         self.sessionkey = sessionkey
+        self.deliverycontext = deliverycontext
         self.note = note
         self.restartdelayms = restartdelayms
     }
@@ -1914,6 +1917,7 @@ public struct ConfigApplyParams: Codable, Sendable {
         case raw
         case basehash = "baseHash"
         case sessionkey = "sessionKey"
+        case deliverycontext = "deliveryContext"
         case note
         case restartdelayms = "restartDelayMs"
     }
@@ -1923,6 +1927,7 @@ public struct ConfigPatchParams: Codable, Sendable {
     public let raw: String
     public let basehash: String?
     public let sessionkey: String?
+    public let deliverycontext: [String: AnyCodable]?
     public let note: String?
     public let restartdelayms: Int?
 
@@ -1930,12 +1935,14 @@ public struct ConfigPatchParams: Codable, Sendable {
         raw: String,
         basehash: String?,
         sessionkey: String?,
+        deliverycontext: [String: AnyCodable]?,
         note: String?,
         restartdelayms: Int?)
     {
         self.raw = raw
         self.basehash = basehash
         self.sessionkey = sessionkey
+        self.deliverycontext = deliverycontext
         self.note = note
         self.restartdelayms = restartdelayms
     }
@@ -1944,6 +1951,7 @@ public struct ConfigPatchParams: Codable, Sendable {
         case raw
         case basehash = "baseHash"
         case sessionkey = "sessionKey"
+        case deliverycontext = "deliveryContext"
         case note
         case restartdelayms = "restartDelayMs"
     }
@@ -4313,17 +4321,20 @@ public struct ChatEvent: Codable, Sendable {
 
 public struct UpdateRunParams: Codable, Sendable {
     public let sessionkey: String?
+    public let deliverycontext: [String: AnyCodable]?
     public let note: String?
     public let restartdelayms: Int?
     public let timeoutms: Int?
 
     public init(
         sessionkey: String?,
+        deliverycontext: [String: AnyCodable]?,
         note: String?,
         restartdelayms: Int?,
         timeoutms: Int?)
     {
         self.sessionkey = sessionkey
+        self.deliverycontext = deliverycontext
         self.note = note
         self.restartdelayms = restartdelayms
         self.timeoutms = timeoutms
@@ -4331,6 +4342,7 @@ public struct UpdateRunParams: Codable, Sendable {
 
     private enum CodingKeys: String, CodingKey {
         case sessionkey = "sessionKey"
+        case deliverycontext = "deliveryContext"
         case note
         case restartdelayms = "restartDelayMs"
         case timeoutms = "timeoutMs"

apps/shared/OpenClawKit/Tools/CanvasA2UI/bootstrap.js

@@ -466,8 +466,10 @@ class OpenClawA2UIHost extends LitElement {
       try {
         // WebKit message handlers support structured objects; Android's JS interface expects strings.
         if (handler === globalThis.openclawCanvasA2UIAction) {
+          // oxlint-disable-next-line unicorn/require-post-message-target-origin -- Native app message handler, not Window.postMessage.
           handler.postMessage(JSON.stringify({ userAction }));
         } else {
+          // oxlint-disable-next-line unicorn/require-post-message-target-origin -- WebKit message handler, not Window.postMessage.
           handler.postMessage({ userAction });
         }
       } catch (e) {

docs/.generated/config-baseline.sha256

@@ -1,4 +1,4 @@
-a962c1d7ddffa15f2333854f77b03da4f6db07fada16f288377ee1daf50afc08  config-baseline.json
-3c8455d44a63d495ad295d2c9d76fed7a190b80344dabaa0e78ba433bf2d253b  config-baseline.core.json
-df55c673a1cdbebc4fe68baaaf9d0d4289313be5034be92f0d510726a086b1d6  config-baseline.channel.json
-3f6fccab66a9abe7e1dd412fb01b13b944ed24edbe09df55ada3323acc7f76fe  config-baseline.plugin.json
+228031f16ad06580bfd137f092d70d03f2796515e723b8b6618ed69d285465fa  config-baseline.json
+bad0a5bb247a62b8fb9ed9fc2b2720eacf3e0913077ac351b5d26ae2723335ad  config-baseline.core.json
+e1f94346a8507ce3dec763b598e79f3bb89ff2e33189ce977cc87d3b05e71c1d  config-baseline.channel.json
+6c19997f1fb2aff4315f2cb9c7d9e299b403fbc0f9e78e3412cc7fe1c655f222  config-baseline.plugin.json

docs/.generated/plugin-sdk-api-baseline.sha256

@@ -1,2 +1,2 @@
-70ecd040dd5815b237eb749db2de98e300fd53630e5b418deea0865b781adf4a  plugin-sdk-api-baseline.json
-7a7b08495662e48d466a1c7cb20d6e81557df05d361de0bebc54369e34dffa35  plugin-sdk-api-baseline.jsonl
+ee16273fa5ad8c5408e9dad8d96fde86dfa666ef8eb44840b78135814ff97173  plugin-sdk-api-baseline.json
+2bd0d5edf23e6a889d6bedb74d0d06411dd7750dac6ebf24971c789f8a69253a  plugin-sdk-api-baseline.jsonl

docs/automation/cron-jobs.md

@@ -43,6 +43,8 @@ together`, and similar hints) and no descendant subagent run is still
   responsible for the final answer, OpenClaw re-prompts once for the actual
   result before delivery.
 
+<a id="maintenance"></a>
+
 Task reconciliation for cron is runtime-owned: an active cron task stays live while the
 cron runtime still tracks that job as running, even if an old child session row still exists.
 Once the runtime stops owning the job and the 5-minute grace window expires, maintenance can

docs/automation/hooks.md

@@ -164,10 +164,14 @@ Enable any bundled hook:
 openclaw hooks enable <hook-name>
 ```
 
+<a id="session-memory"></a>
+
 ### session-memory details
 
 Extracts the last 15 user/assistant messages, generates a descriptive filename slug via LLM, and saves to `<workspace>/memory/YYYY-MM-DD-slug.md`. Requires `workspace.dir` to be configured.
 
+<a id="bootstrap-extra-files"></a>
+
 ### bootstrap-extra-files config
 
 ```json
@@ -187,6 +191,18 @@ Extracts the last 15 user/assistant messages, generates a descriptive filename s
 
 Paths resolve relative to workspace. Only recognized bootstrap basenames are loaded (`AGENTS.md`, `SOUL.md`, `TOOLS.md`, `IDENTITY.md`, `USER.md`, `HEARTBEAT.md`, `BOOTSTRAP.md`, `MEMORY.md`).
 
+<a id="command-logger"></a>
+
+### command-logger details
+
+Logs every slash command to `~/.openclaw/logs/commands.log`.
+
+<a id="boot-md"></a>
+
+### boot-md details
+
+Runs `BOOT.md` from the active workspace when the gateway starts.
+
 ## Plugin hooks
 
 Plugins can register hooks through the Plugin SDK for deeper integration: intercepting tool calls, modifying prompts, controlling message flow, and more. The Plugin SDK exposes 28 hooks covering model resolution, agent lifecycle, message flow, tool execution, subagent coordination, and gateway lifecycle.

docs/ci.md

@@ -12,24 +12,25 @@ The CI runs on every push to `main` and every pull request. It uses smart scopin
 
 ## Job Overview
 
-| Job                      | Purpose                                                                                  | When it runs                        |
-| ------------------------ | ---------------------------------------------------------------------------------------- | ----------------------------------- |
-| `preflight`              | Detect docs-only changes, changed scopes, changed extensions, and build the CI manifest  | Always on non-draft pushes and PRs  |
-| `security-fast`          | Private key detection, workflow audit via `zizmor`, production dependency audit          | Always on non-draft pushes and PRs  |
-| `build-artifacts`        | Build `dist/` and the Control UI once, upload reusable artifacts for downstream jobs     | Node-relevant changes               |
-| `checks-fast-core`       | Fast Linux correctness lanes such as bundled/plugin-contract/protocol checks             | Node-relevant changes               |
-| `checks-fast-extensions` | Aggregate the extension shard lanes after `checks-fast-extensions-shard` completes       | Node-relevant changes               |
-| `extension-fast`         | Focused tests for only the changed bundled plugins                                       | When extension changes are detected |
-| `check`                  | Main local gate in CI: `pnpm check` plus `pnpm build:strict-smoke`                       | Node-relevant changes               |
-| `check-additional`       | Architecture, boundary, import-cycle guards plus the gateway watch regression harness    | Node-relevant changes               |
-| `build-smoke`            | Built-CLI smoke tests and startup-memory smoke                                           | Node-relevant changes               |
-| `checks`                 | Heavier Linux Node lanes: full tests, channel tests, and push-only Node 22 compatibility | Node-relevant changes               |
-| `check-docs`             | Docs formatting, lint, and broken-link checks                                            | Docs changed                        |
-| `skills-python`          | Ruff + pytest for Python-backed skills                                                   | Python-skill-relevant changes       |
-| `checks-windows`         | Windows-specific test lanes                                                              | Windows-relevant changes            |
-| `macos-node`             | macOS TypeScript test lane using the shared built artifacts                              | macOS-relevant changes              |
-| `macos-swift`            | Swift lint, build, and tests for the macOS app                                           | macOS-relevant changes              |
-| `android`                | Android build and test matrix                                                            | Android-relevant changes            |
+| Job                      | Purpose                                                                                 | When it runs                        |
+| ------------------------ | --------------------------------------------------------------------------------------- | ----------------------------------- |
+| `preflight`              | Detect docs-only changes, changed scopes, changed extensions, and build the CI manifest | Always on non-draft pushes and PRs  |
+| `security-fast`          | Private key detection, workflow audit via `zizmor`, production dependency audit         | Always on non-draft pushes and PRs  |
+| `build-artifacts`        | Build `dist/` and the Control UI once, upload reusable artifacts for downstream jobs    | Node-relevant changes               |
+| `checks-fast-core`       | Fast Linux correctness lanes such as bundled/plugin-contract/protocol checks            | Node-relevant changes               |
+| `checks-node-extensions` | Full bundled-plugin test shards across the extension suite                              | Node-relevant changes               |
+| `checks-node-core-test`  | Core Node test shards, excluding channel, bundled, contract, and extension lanes        | Node-relevant changes               |
+| `extension-fast`         | Focused tests for only the changed bundled plugins                                      | When extension changes are detected |
+| `check`                  | Main local gate in CI: `pnpm check` plus `pnpm build:strict-smoke`                      | Node-relevant changes               |
+| `check-additional`       | Architecture, boundary, import-cycle guards plus the gateway watch regression harness   | Node-relevant changes               |
+| `build-smoke`            | Built-CLI smoke tests and startup-memory smoke                                          | Node-relevant changes               |
+| `checks`                 | Remaining Linux Node lanes: channel tests and push-only Node 22 compatibility           | Node-relevant changes               |
+| `check-docs`             | Docs formatting, lint, and broken-link checks                                           | Docs changed                        |
+| `skills-python`          | Ruff + pytest for Python-backed skills                                                  | Python-skill-relevant changes       |
+| `checks-windows`         | Windows-specific test lanes                                                             | Windows-relevant changes            |
+| `macos-node`             | macOS TypeScript test lane using the shared built artifacts                             | macOS-relevant changes              |
+| `macos-swift`            | Swift lint, build, and tests for the macOS app                                          | macOS-relevant changes              |
+| `android`                | Android build and test matrix                                                           | Android-relevant changes            |
 
 ## Fail-Fast Order
 
@@ -38,7 +39,7 @@ Jobs are ordered so cheap checks fail before expensive ones run:
 1. `preflight` decides which lanes exist at all. The `docs-scope` and `changed-scope` logic are steps inside this job, not standalone jobs.
 2. `security-fast`, `check`, `check-additional`, `check-docs`, and `skills-python` fail quickly without waiting on the heavier artifact and platform matrix jobs.
 3. `build-artifacts` overlaps with the fast Linux lanes so downstream consumers can start as soon as the shared build is ready.
-4. Heavier platform and runtime lanes fan out after that: `checks-fast-core`, `checks-fast-extensions`, `extension-fast`, `checks`, `checks-windows`, `macos-node`, `macos-swift`, and `android`.
+4. Heavier platform and runtime lanes fan out after that: `checks-fast-core`, `checks-node-extensions`, `checks-node-core-test`, `extension-fast`, `checks`, `checks-windows`, `macos-node`, `macos-swift`, and `android`.
 
 Scope logic lives in `scripts/ci-changed-scope.mjs` and is covered by unit tests in `src/scripts/ci-changed-scope.test.ts`.
 The separate `install-smoke` workflow reuses the same scope script through its own `preflight` job. It computes `run_install_smoke` from the narrower changed-smoke signal, so Docker/install smoke only runs for install, packaging, and container-relevant changes.

docs/cli/agents.md

@@ -37,7 +37,7 @@ Use routing bindings to pin inbound channel traffic to a specific agent.
 If you also want different visible skills per agent, configure
 `agents.defaults.skills` and `agents.list[].skills` in `openclaw.json`. See
 [Skills config](/tools/skills-config) and
-[Configuration Reference](/gateway/configuration-reference#agentsdefaultsskills).
+[Configuration Reference](/gateway/configuration-reference#agents-defaults-skills).
 
 List bindings:
 

docs/cli/devices.md

@@ -49,8 +49,10 @@ openclaw devices clear --yes --pending --json
 
 ### `openclaw devices approve [requestId] [--latest]`
 
-Approve a pending device pairing request. If `requestId` is omitted, OpenClaw
-automatically approves the most recent pending request.
+Approve a pending device pairing request by exact `requestId`. If `requestId`
+is omitted or `--latest` is passed, OpenClaw only prints the selected pending
+request and exits; rerun approval with the exact request ID after verifying
+the details.
 
 Note: if a device retries pairing with changed auth details (role/scopes/public
 key), OpenClaw supersedes the previous pending entry and issues a new
@@ -126,7 +128,7 @@ Pass `--token` or `--password` explicitly. Missing explicit credentials is an er
   `operator.admin`.
 - `devices clear` is intentionally gated by `--yes`.
 - If pairing scope is unavailable on local loopback (and no explicit `--url` is passed), list/approve can use a local pairing fallback.
-- `devices approve` picks the newest pending request automatically when you omit `requestId` or pass `--latest`.
+- `devices approve` requires an explicit request ID before minting tokens; omitting `requestId` or passing `--latest` only previews the newest pending request.
 
 ## Token drift recovery checklist
 

docs/cli/index.md

@@ -852,7 +852,7 @@ Subcommands:
 Notes:
 
 - `devices list` and `devices approve` can fall back to local pairing files on local loopback when direct pairing scope is unavailable.
-- `devices approve` auto-selects the newest pending request when no `requestId` is passed or `--latest` is set.
+- `devices approve` requires an explicit request ID before minting tokens; omitting `requestId` or passing `--latest` only previews the newest pending request.
 - Stored-token reconnects reuse the token's cached approved scopes; explicit
   `devices rotate --scope ...` updates that stored scope set for future
   cached-token reconnects.

docs/concepts/active-memory.md

@@ -57,7 +57,7 @@ available.
 After that, restart the gateway:
 
 ```bash
-node scripts/run-node.mjs gateway --profile dev
+openclaw gateway
 ```
 
 To inspect it live in a conversation:
@@ -102,7 +102,7 @@ Start with this in `openclaw.json`:
 Then restart the gateway:
 
 ```bash
-node scripts/run-node.mjs gateway --profile dev
+openclaw gateway
 ```
 
 What this means:

docs/concepts/model-providers.md

@@ -50,6 +50,13 @@ For model selection rules, see [/concepts/models](/concepts/models).
   family, transcript/tooling quirks, transport/cache hints). It is not the
   same as the [public capability model](/plugins/architecture#public-capability-model)
   which describes what a plugin registers (text inference, speech, etc.).
+- The bundled `codex` provider is paired with the bundled Codex agent harness.
+  Use `codex/gpt-*` when you want Codex-owned login, model discovery, native
+  thread resume, and app-server execution. Plain `openai/gpt-*` refs continue
+  to use the OpenAI provider and the normal OpenClaw provider transport.
+  Codex-only deployments can disable automatic PI fallback with
+  `agents.defaults.embeddedHarness.fallback: "none"`; see
+  [Codex Harness](/plugins/codex-harness).
 
 ## Plugin-owned provider behavior
 

docs/concepts/qa-e2e-automation.md

@@ -52,6 +52,47 @@ pnpm qa:lab:watch
 rebuilds that bundle on change, and the browser auto-reloads when the QA Lab
 asset hash changes.
 
+For a transport-real Matrix smoke lane, run:
+
+```bash
+pnpm openclaw qa matrix
+```
+
+That lane provisions a disposable Tuwunel homeserver in Docker, registers
+temporary driver, SUT, and observer users, creates one private room, then runs
+the real Matrix plugin inside a QA gateway child. The live transport lane keeps
+the child config scoped to the transport under test, so Matrix runs without
+`qa-channel` in the child config.
+
+For a transport-real Telegram smoke lane, run:
+
+```bash
+pnpm openclaw qa telegram
+```
+
+That lane targets one real private Telegram group instead of provisioning a
+disposable server. It requires `OPENCLAW_QA_TELEGRAM_GROUP_ID`,
+`OPENCLAW_QA_TELEGRAM_DRIVER_BOT_TOKEN`, and
+`OPENCLAW_QA_TELEGRAM_SUT_BOT_TOKEN`, plus two distinct bots in the same
+private group. The SUT bot must have a Telegram username, and bot-to-bot
+observation works best when both bots have Bot-to-Bot Communication Mode
+enabled in `@BotFather`.
+
+Live transport lanes now share one smaller contract instead of each inventing
+their own scenario list shape:
+
+`qa-channel` remains the broad synthetic product-behavior suite and is not part
+of the live transport coverage matrix.
+
+| Lane     | Canary | Mention gating | Allowlist block | Top-level reply | Restart resume | Thread follow-up | Thread isolation | Reaction observation | Help command |
+| -------- | ------ | -------------- | --------------- | --------------- | -------------- | ---------------- | ---------------- | -------------------- | ------------ |
+| Matrix   | x      | x              | x               | x               | x              | x                | x                | x                    |              |
+| Telegram | x      |                |                 |                 |                |                  |                  |                      | x            |
+
+This keeps `qa-channel` as the broad product-behavior suite while Matrix,
+Telegram, and future live transports share one explicit transport-contract
+checklist.
+
 For a disposable Linux VM lane without bringing Docker into the QA path, run:
 
 ```bash

docs/docs.json

@@ -1113,6 +1113,7 @@
                   "tools/plugin",
                   "plugins/community",
                   "plugins/bundles",
+                  "plugins/codex-harness",
                   "plugins/webhooks",
                   "plugins/voice-call",
                   {
@@ -1130,6 +1131,7 @@
                       "plugins/sdk-overview",
                       "plugins/sdk-entrypoints",
                       "plugins/sdk-runtime",
+                      "plugins/sdk-agent-harness",
                       "plugins/sdk-setup",
                       "plugins/sdk-testing",
                       "plugins/manifest",

docs/gateway/cli-backends.md

@@ -263,6 +263,31 @@ CLI backend defaults are now part of the plugin surface:
 - Backend-specific config cleanup stays plugin-owned through the optional
   `normalizeConfig` hook.
 
+Plugins that need tiny prompt/message compatibility shims can declare
+bidirectional text transforms without replacing a provider or CLI backend:
+
+```typescript
+api.registerTextTransforms({
+  input: [
+    { from: /red basket/g, to: "blue basket" },
+    { from: /paper ticket/g, to: "digital ticket" },
+    { from: /left shelf/g, to: "right shelf" },
+  ],
+  output: [
+    { from: /blue basket/g, to: "red basket" },
+    { from: /digital ticket/g, to: "paper ticket" },
+    { from: /right shelf/g, to: "left shelf" },
+  ],
+});
+```
+
+`input` rewrites the system prompt and user prompt passed to the CLI. `output`
+rewrites streamed assistant deltas and parsed final text before OpenClaw handles
+its own control markers and channel delivery.
+
+For CLIs that emit Claude Code stream-json compatible JSONL, set
+`jsonlDialect: "claude-stream-json"` on that backend's config.
+
 ## Bundle MCP overlays
 
 CLI backends do **not** receive OpenClaw tool calls directly, but a backend can

docs/gateway/configuration-reference.md

@@ -1053,6 +1053,10 @@ Time format in system prompt. Default: `auto` (OS preference).
         fallbacks: ["openai/gpt-5.4-mini"],
       },
       params: { cacheRetention: "long" }, // global default provider params
+      embeddedHarness: {
+        runtime: "auto", // auto | pi | registered harness id, e.g. codex
+        fallback: "pi", // pi | none
+      },
       pdfMaxBytesMb: 10,
       pdfMaxPages: 20,
       thinkingDefault: "low",
@@ -1100,9 +1104,37 @@ Time format in system prompt. Default: `auto` (OS preference).
 - `models`: the configured model catalog and allowlist for `/model`. Each entry can include `alias` (shortcut) and `params` (provider-specific, for example `temperature`, `maxTokens`, `cacheRetention`, `context1m`).
 - `params`: global default provider parameters applied to all models. Set at `agents.defaults.params` (e.g. `{ cacheRetention: "long" }`).
 - `params` merge precedence (config): `agents.defaults.params` (global base) is overridden by `agents.defaults.models["provider/model"].params` (per-model), then `agents.list[].params` (matching agent id) overrides by key. See [Prompt Caching](/reference/prompt-caching) for details.
+- `embeddedHarness`: default low-level embedded agent runtime policy. Use `runtime: "auto"` to let registered plugin harnesses claim supported models, `runtime: "pi"` to force the built-in PI harness, or a registered harness id such as `runtime: "codex"`. Set `fallback: "none"` to disable automatic PI fallback.
 - Config writers that mutate these fields (for example `/models set`, `/models set-image`, and fallback add/remove commands) save canonical object form and preserve existing fallback lists when possible.
 - `maxConcurrent`: max parallel agent runs across sessions (each session still serialized). Default: 4.
 
+### `agents.defaults.embeddedHarness`
+
+`embeddedHarness` controls which low-level executor runs embedded agent turns.
+Most deployments should keep the default `{ runtime: "auto", fallback: "pi" }`.
+Use it when a trusted plugin provides a native harness, such as the bundled
+Codex app-server harness.
+
+```json5
+{
+  agents: {
+    defaults: {
+      model: "codex/gpt-5.4",
+      embeddedHarness: {
+        runtime: "codex",
+        fallback: "none",
+      },
+    },
+  },
+}
+```
+
+- `runtime`: `"auto"`, `"pi"`, or a registered plugin harness id. The bundled Codex plugin registers `codex`.
+- `fallback`: `"pi"` or `"none"`. `"pi"` keeps the built-in PI harness as the compatibility fallback. `"none"` makes missing or unsupported plugin harness selection fail instead of silently using PI.
+- Environment overrides: `OPENCLAW_AGENT_RUNTIME=<id|auto|pi>` overrides `runtime`; `OPENCLAW_AGENT_HARNESS_FALLBACK=none` disables PI fallback for that process.
+- For Codex-only deployments, set `model: "codex/gpt-5.4"`, `embeddedHarness.runtime: "codex"`, and `embeddedHarness.fallback: "none"`.
+- This only controls the embedded chat harness. Media generation, vision, PDF, music, video, and TTS still use their provider/model settings.
+
 **Built-in alias shorthands** (only apply when the model is in `agents.defaults.models`):
 
 | Alias               | Model                                  |
@@ -1192,6 +1224,7 @@ Periodic heartbeat runs.
         prompt: "Read HEARTBEAT.md if it exists...",
         ackMaxChars: 300,
         suppressToolErrorWarnings: false,
+        timeoutSeconds: 45,
       },
     },
   },
@@ -1201,6 +1234,7 @@ Periodic heartbeat runs.
 - `every`: duration string (ms/s/m/h). Default: `30m` (API-key auth) or `1h` (OAuth auth). Set to `0m` to disable.
 - `includeSystemPromptSection`: when false, omits the Heartbeat section from the system prompt and skips `HEARTBEAT.md` injection into bootstrap context. Default: `true`.
 - `suppressToolErrorWarnings`: when true, suppresses tool error warning payloads during heartbeat runs.
+- `timeoutSeconds`: maximum time in seconds allowed for a heartbeat agent turn before it is aborted. Leave unset to use `agents.defaults.timeoutSeconds`.
 - `directPolicy`: direct/DM delivery policy. `allow` (default) permits direct-target delivery. `block` suppresses direct-target delivery and emits `reason=dm-blocked`.
 - `lightContext`: when true, heartbeat runs use lightweight bootstrap context and keep only `HEARTBEAT.md` from workspace bootstrap files.
 - `isolatedSession`: when true, each heartbeat runs in a fresh session with no prior conversation history. Same isolation pattern as cron `sessionTarget: "isolated"`. Reduces per-heartbeat token cost from ~100K to ~2-5K tokens.
@@ -1583,6 +1617,7 @@ scripts/sandbox-browser-setup.sh   # optional browser image
         thinkingDefault: "high", // per-agent thinking level override
         reasoningDefault: "on", // per-agent reasoning visibility override
         fastModeDefault: false, // per-agent fast mode override
+        embeddedHarness: { runtime: "auto", fallback: "pi" },
         params: { cacheRetention: "none" }, // overrides matching defaults.models params by key
         skills: ["docs-search"], // replaces agents.defaults.skills when set
         identity: {
@@ -1623,6 +1658,7 @@ scripts/sandbox-browser-setup.sh   # optional browser image
 - `thinkingDefault`: optional per-agent default thinking level (`off | minimal | low | medium | high | xhigh | adaptive`). Overrides `agents.defaults.thinkingDefault` for this agent when no per-message or session override is set.
 - `reasoningDefault`: optional per-agent default reasoning visibility (`on | off | stream`). Applies when no per-message or session reasoning override is set.
 - `fastModeDefault`: optional per-agent default for fast mode (`true | false`). Applies when no per-message or session fast-mode override is set.
+- `embeddedHarness`: optional per-agent low-level harness policy override. Use `{ runtime: "codex", fallback: "none" }` to make one agent Codex-only while other agents keep the default PI fallback.
 - `runtime`: optional per-agent runtime descriptor. Use `type: "acp"` with `runtime.acp` defaults (`agent`, `backend`, `mode`, `cwd`) when the agent should default to ACP harness sessions.
 - `identity.avatar`: workspace-relative path, `http(s)` URL, or `data:` URI.
 - `identity` derives defaults: `ackReaction` from `emoji`, `mentionPatterns` from `name`/`emoji`.
@@ -2299,7 +2335,7 @@ Notes:
 
 ### `tools.experimental`
 
-Experimental built-in tool flags. Default off unless a runtime-specific auto-enable rule applies.
+Experimental built-in tool flags. Default off unless a strict-agentic GPT-5 auto-enable rule applies.
 
 ```json5
 {
@@ -2314,7 +2350,7 @@ Experimental built-in tool flags. Default off unless a runtime-specific auto-ena
 Notes:
 
 - `planTool`: enables the structured `update_plan` tool for non-trivial multi-step work tracking.
-- Default: `false` for non-OpenAI providers. OpenAI and OpenAI Codex runs auto-enable it when unset; set `false` to disable that auto-enable.
+- Default: `false` unless `agents.defaults.embeddedPi.executionContract` (or a per-agent override) is set to `"strict-agentic"` for an OpenAI or OpenAI Codex GPT-5-family run. Set `true` to force the tool on outside that scope, or `false` to keep it off even for strict-agentic GPT-5 runs.
 - When enabled, the system prompt also adds usage guidance so the model only uses it for substantial work and keeps at most one step `in_progress`.
 
 ### `agents.defaults.subagents`
@@ -2758,7 +2794,7 @@ See [Plugins](/tools/plugin).
     evaluateEnabled: true,
     defaultProfile: "user",
     ssrfPolicy: {
-      dangerouslyAllowPrivateNetwork: true, // default trusted-network mode
+      // dangerouslyAllowPrivateNetwork: true, // opt in only for trusted private-network access
       // allowPrivateNetwork: true, // legacy alias
       // hostnameAllowlist: ["*.example.com", "example.com"],
       // allowedHostnames: ["localhost"],
@@ -2786,8 +2822,8 @@ See [Plugins](/tools/plugin).
 ```
 
 - `evaluateEnabled: false` disables `act:evaluate` and `wait --fn`.
-- `ssrfPolicy.dangerouslyAllowPrivateNetwork` defaults to `true` when unset (trusted-network model).
-- Set `ssrfPolicy.dangerouslyAllowPrivateNetwork: false` for strict public-only browser navigation.
+- `ssrfPolicy.dangerouslyAllowPrivateNetwork` is disabled when unset, so browser navigation stays strict by default.
+- Set `ssrfPolicy.dangerouslyAllowPrivateNetwork: true` only when you intentionally trust private-network browser navigation.
 - In strict mode, remote CDP profile endpoints (`profiles.*.cdpUrl`) are subject to the same private-network blocking during reachability/discovery checks.
 - `ssrfPolicy.allowPrivateNetwork` remains supported as a legacy alias.
 - In strict mode, use `ssrfPolicy.hostnameAllowlist` and `ssrfPolicy.allowedHostnames` for explicit exceptions.

docs/gateway/configuration.md

@@ -224,7 +224,7 @@ When validation fails:
     - Omit `agents.list[].skills` to inherit the defaults.
     - Set `agents.list[].skills: []` for no skills.
     - See [Skills](/tools/skills), [Skills config](/tools/skills-config), and
-      the [Configuration Reference](/gateway/configuration-reference#agentsdefaultsskills).
+      the [Configuration Reference](/gateway/configuration-reference#agents-defaults-skills).
 
   </Accordion>
 

docs/gateway/heartbeat.md

@@ -146,6 +146,7 @@ Example: two agents, only the second agent runs heartbeats.
           every: "1h",
           target: "whatsapp",
           to: "+15551234567",
+          timeoutSeconds: 45,
           prompt: "Read HEARTBEAT.md if it exists (workspace context). Follow it strictly. Do not infer or repeat old tasks from prior chats. If nothing needs attention, reply HEARTBEAT_OK.",
         },
       },

docs/gateway/security/index.md

@@ -13,7 +13,7 @@ OpenClaw is **not** a hostile multi-tenant security boundary for multiple advers
 If you need mixed-trust or adversarial-user operation, split trust boundaries (separate gateway + credentials, ideally separate OS users/hosts).
 </Warning>
 
-**On this page:** [Trust model](#scope-first-personal-assistant-security-model) | [Quick audit](#quick-check-openclaw-security-audit) | [Hardened baseline](#hardened-baseline-in-60-seconds) | [DM access model](#dm-access-model-pairing--allowlist--open--disabled) | [Configuration hardening](#configuration-hardening-examples) | [Incident response](#incident-response)
+**On this page:** [Trust model](#scope-first-personal-assistant-security-model) | [Quick audit](#quick-check-openclaw-security-audit) | [Hardened baseline](#hardened-baseline-in-60-seconds) | [DM access model](#dm-access-model-pairing-allowlist-open-disabled) | [Configuration hardening](#configuration-hardening-examples) | [Incident response](#incident-response)
 
 ## Scope first: personal assistant security model
 
@@ -187,7 +187,7 @@ Allowlists gate triggers and command authorization. The `contextVisibility` sett
 - `contextVisibility: "allowlist"` filters supplemental context to senders allowed by the active allowlist checks.
 - `contextVisibility: "allowlist_quote"` behaves like `allowlist`, but still keeps one explicit quoted reply.
 
-Set `contextVisibility` per channel or per room/conversation. See [Group Chats](/channels/groups#context-visibility) for setup details.
+Set `contextVisibility` per channel or per room/conversation. See [Group Chats](/channels/groups#context-visibility-and-allowlists) for setup details.
 
 Advisory triage guidance:
 
@@ -579,6 +579,8 @@ Plugins run **in-process** with the Gateway. Treat them as trusted code:
 
 Details: [Plugins](/tools/plugin)
 
+<a id="dm-access-model-pairing-allowlist-open-disabled"></a>
+
 ## DM access model (pairing / allowlist / open / disabled)
 
 All current DM-capable channels support a DM policy (`dmPolicy` or `*.dm.policy`) that gates inbound DMs **before** the message is processed:
@@ -1149,13 +1151,13 @@ access those accounts and data. Treat browser profiles as **sensitive state**:
 - Disable browser proxy routing when you don’t need it (`gateway.nodes.browser.mode="off"`).
 - Chrome MCP existing-session mode is **not** “safer”; it can act as you in whatever that host Chrome profile can reach.
 
-### Browser SSRF policy (trusted-network default)
+### Browser SSRF policy (strict by default)
 
-OpenClaw’s browser network policy defaults to the trusted-operator model: private/internal destinations are allowed unless you explicitly disable them.
+OpenClaw’s browser navigation policy is strict by default: private/internal destinations stay blocked unless you explicitly opt in.
 
-- Default: `browser.ssrfPolicy.dangerouslyAllowPrivateNetwork: true` (implicit when unset).
+- Default: `browser.ssrfPolicy.dangerouslyAllowPrivateNetwork` is unset, so browser navigation keeps private/internal/special-use destinations blocked.
 - Legacy alias: `browser.ssrfPolicy.allowPrivateNetwork` is still accepted for compatibility.
-- Strict mode: set `browser.ssrfPolicy.dangerouslyAllowPrivateNetwork: false` to block private/internal/special-use destinations by default.
+- Opt-in mode: set `browser.ssrfPolicy.dangerouslyAllowPrivateNetwork: true` to allow private/internal/special-use destinations.
 - In strict mode, use `hostnameAllowlist` (patterns like `*.example.com`) and `allowedHostnames` (exact host exceptions, including blocked names like `localhost`) for explicit exceptions.
 - Navigation is checked before request and best-effort re-checked on the final `http(s)` URL after navigation to reduce redirect-based pivots.
 

docs/gateway/troubleshooting.md

@@ -111,7 +111,7 @@ Fix options:
 Related:
 
 - [/gateway/local-models](/gateway/local-models)
-- [/gateway/configuration#models](/gateway/configuration#models)
+- [/gateway/configuration](/gateway/configuration)
 - [/gateway/configuration-reference#openai-compatible-endpoints](/gateway/configuration-reference#openai-compatible-endpoints)
 
 ## No replies

docs/help/testing.md

@@ -65,6 +65,27 @@ These commands sit beside the main test suites when you need QA-lab realism:
     `.artifacts/qa-e2e/...`.
 - `pnpm qa:lab:up`
   - Starts the Docker-backed QA site for operator-style QA work.
+- `pnpm openclaw qa matrix`
+  - Runs the Matrix live QA lane against a disposable Docker-backed Tuwunel homeserver.
+  - Provisions three temporary Matrix users (`driver`, `sut`, `observer`) plus one private room, then starts a QA gateway child with the real Matrix plugin as the SUT transport.
+  - Uses the pinned stable Tuwunel image `ghcr.io/matrix-construct/tuwunel:v1.5.1` by default. Override with `OPENCLAW_QA_MATRIX_TUWUNEL_IMAGE` when you need to test a different image.
+  - Writes a Matrix QA report, summary, and observed-events artifact under `.artifacts/qa-e2e/...`.
+- `pnpm openclaw qa telegram`
+  - Runs the Telegram live QA lane against a real private group using the driver and SUT bot tokens from env.
+  - Requires `OPENCLAW_QA_TELEGRAM_GROUP_ID`, `OPENCLAW_QA_TELEGRAM_DRIVER_BOT_TOKEN`, and `OPENCLAW_QA_TELEGRAM_SUT_BOT_TOKEN`. The group id must be the numeric Telegram chat id.
+  - Requires two distinct bots in the same private group, with the SUT bot exposing a Telegram username.
+  - For stable bot-to-bot observation, enable Bot-to-Bot Communication Mode in `@BotFather` for both bots and ensure the driver bot can observe group bot traffic.
+  - Writes a Telegram QA report, summary, and observed-messages artifact under `.artifacts/qa-e2e/...`.
+
+Live transport lanes share one standard contract so new transports do not drift:
+
+`qa-channel` remains the broad synthetic QA suite and is not part of the live
+transport coverage matrix.
+
+| Lane     | Canary | Mention gating | Allowlist block | Top-level reply | Restart resume | Thread follow-up | Thread isolation | Reaction observation | Help command |
+| -------- | ------ | -------------- | --------------- | --------------- | -------------- | ---------------- | ---------------- | -------------------- | ------------ |
+| Matrix   | x      | x              | x               | x               | x              | x                | x                | x                    |              |
+| Telegram | x      |                |                 |                 |                |                  |                  |                      | x            |
 
 ## Test suites (what runs where)
 
@@ -88,7 +109,7 @@ Think of the suites as “increasing realism” (and increasing flakiness/cost):
   - `pnpm test --watch` still uses the native root `vitest.config.ts` project graph, because a multi-shard watch loop is not practical.
   - `pnpm test`, `pnpm test:watch`, and `pnpm test:perf:imports` route explicit file/directory targets through scoped lanes first, so `pnpm test extensions/discord/src/monitor/message-handler.preflight.test.ts` avoids paying the full root project startup tax.
   - `pnpm test:changed` expands changed git paths into the same scoped lanes when the diff only touches routable source/test files; config/setup edits still fall back to the broad root-project rerun.
-  - Selected `plugin-sdk` and `commands` tests also route through dedicated light lanes that skip `test/setup-openclaw-runtime.ts`; stateful/runtime-heavy files stay on the existing lanes.
+  - Import-light unit tests from agents, commands, plugins, auto-reply helpers, `plugin-sdk`, and similar pure utility areas route through the `unit-fast` lane, which skips `test/setup-openclaw-runtime.ts`; stateful/runtime-heavy files stay on the existing lanes.
   - Selected `plugin-sdk` and `commands` helper source files also map changed-mode runs to explicit sibling tests in those light lanes, so helper edits avoid rerunning the full heavy suite for that directory.
   - `auto-reply` now has three dedicated buckets: top-level core helpers, top-level `reply.*` integration tests, and the `src/auto-reply/reply/**` subtree. This keeps the heaviest reply harness work off the cheap status/chunk/token tests.
 - Embedded runner note:
@@ -390,6 +411,58 @@ Docker notes:
 - It sources `~/.profile`, stages the matching CLI auth material into the container, installs `acpx` into a writable npm prefix, then installs the requested live CLI (`@anthropic-ai/claude-code`, `@openai/codex`, or `@google/gemini-cli`) if missing.
 - Inside Docker, the runner sets `OPENCLAW_LIVE_ACP_BIND_ACPX_COMMAND=$HOME/.npm-global/bin/acpx` so acpx keeps provider env vars from the sourced profile available to the child harness CLI.
 
+## Live: Codex app-server harness smoke
+
+- Goal: validate the plugin-owned Codex harness through the normal gateway
+  `agent` method:
+  - load the bundled `codex` plugin
+  - select `OPENCLAW_AGENT_RUNTIME=codex`
+  - send a first gateway agent turn to `codex/gpt-5.4`
+  - send a second turn to the same OpenClaw session and verify the app-server
+    thread can resume
+  - run `/codex status` and `/codex models` through the same gateway command
+    path
+- Test: `src/gateway/gateway-codex-harness.live.test.ts`
+- Enable: `OPENCLAW_LIVE_CODEX_HARNESS=1`
+- Default model: `codex/gpt-5.4`
+- Optional image probe: `OPENCLAW_LIVE_CODEX_HARNESS_IMAGE_PROBE=1`
+- Optional MCP/tool probe: `OPENCLAW_LIVE_CODEX_HARNESS_MCP_PROBE=1`
+- The smoke sets `OPENCLAW_AGENT_HARNESS_FALLBACK=none` so a broken Codex
+  harness cannot pass by silently falling back to PI.
+- Auth: `OPENAI_API_KEY` from the shell/profile, plus optional copied
+  `~/.codex/auth.json` and `~/.codex/config.toml`
+
+Local recipe:
+
+```bash
+source ~/.profile
+OPENCLAW_LIVE_CODEX_HARNESS=1 \
+  OPENCLAW_LIVE_CODEX_HARNESS_IMAGE_PROBE=1 \
+  OPENCLAW_LIVE_CODEX_HARNESS_MCP_PROBE=1 \
+  OPENCLAW_LIVE_CODEX_HARNESS_MODEL=codex/gpt-5.4 \
+  pnpm test:live -- src/gateway/gateway-codex-harness.live.test.ts
+```
+
+Docker recipe:
+
+```bash
+source ~/.profile
+pnpm test:docker:live-codex-harness
+```
+
+Docker notes:
+
+- The Docker runner lives at `scripts/test-live-codex-harness-docker.sh`.
+- It sources the mounted `~/.profile`, passes `OPENAI_API_KEY`, copies Codex CLI
+  auth files when present, installs `@openai/codex` into a writable mounted npm
+  prefix, stages the source tree, then runs only the Codex-harness live test.
+- Docker enables the image and MCP/tool probes by default. Set
+  `OPENCLAW_LIVE_CODEX_HARNESS_IMAGE_PROBE=0` or
+  `OPENCLAW_LIVE_CODEX_HARNESS_MCP_PROBE=0` when you need a narrower debug run.
+- Docker also exports `OPENCLAW_AGENT_HARNESS_FALLBACK=none`, matching the live
+  test config so `openai-codex/*` or PI fallback cannot hide a Codex harness
+  regression.
+
 ### Recommended live recipes
 
 Narrow, explicit allowlists are fastest and least flaky:
@@ -618,6 +691,7 @@ The live-model Docker runners also bind-mount only the needed CLI auth homes (or
 - Direct models: `pnpm test:docker:live-models` (script: `scripts/test-live-models-docker.sh`)
 - ACP bind smoke: `pnpm test:docker:live-acp-bind` (script: `scripts/test-live-acp-bind-docker.sh`)
 - CLI backend smoke: `pnpm test:docker:live-cli-backend` (script: `scripts/test-live-cli-backend-docker.sh`)
+- Codex app-server harness smoke: `pnpm test:docker:live-codex-harness` (script: `scripts/test-live-codex-harness-docker.sh`)
 - Gateway + dev agent: `pnpm test:docker:live-gateway` (script: `scripts/test-live-gateway-models-docker.sh`)
 - Open WebUI live smoke: `pnpm test:docker:openwebui` (script: `scripts/e2e/openwebui-docker.sh`)
 - Onboarding wizard (TTY, full scaffolding): `pnpm test:docker:onboard` (script: `scripts/e2e/onboard-docker.sh`)

docs/help/troubleshooting.md

@@ -251,18 +251,19 @@ flowchart TD
 
     Common log signatures:
 
-- `cron: scheduler disabled; jobs will not run automatically` → cron is disabled.
-- `heartbeat skipped` with `reason=quiet-hours` → outside configured active hours.
-- `heartbeat skipped` with `reason=empty-heartbeat-file` → `HEARTBEAT.md` exists but only contains blank/header-only scaffolding.
-- `heartbeat skipped` with `reason=no-tasks-due` → `HEARTBEAT.md` task mode is active but none of the task intervals are due yet.
-- `heartbeat skipped` with `reason=alerts-disabled` → all heartbeat visibility is disabled (`showOk`, `showAlerts`, and `useIndicator` are all off).
-- `requests-in-flight` → main lane busy; heartbeat wake was deferred. - `unknown accountId` → heartbeat delivery target account does not exist.
+    - `cron: scheduler disabled; jobs will not run automatically` → cron is disabled.
+    - `heartbeat skipped` with `reason=quiet-hours` → outside configured active hours.
+    - `heartbeat skipped` with `reason=empty-heartbeat-file` → `HEARTBEAT.md` exists but only contains blank/header-only scaffolding.
+    - `heartbeat skipped` with `reason=no-tasks-due` → `HEARTBEAT.md` task mode is active but none of the task intervals are due yet.
+    - `heartbeat skipped` with `reason=alerts-disabled` → all heartbeat visibility is disabled (`showOk`, `showAlerts`, and `useIndicator` are all off).
+    - `requests-in-flight` → main lane busy; heartbeat wake was deferred.
+    - `unknown accountId` → heartbeat delivery target account does not exist.
 
-      Deep pages:
+    Deep pages:
 
-      - [/gateway/troubleshooting#cron-and-heartbeat-delivery](/gateway/troubleshooting#cron-and-heartbeat-delivery)
-      - [/automation/cron-jobs#troubleshooting](/automation/cron-jobs#troubleshooting)
-      - [/gateway/heartbeat](/gateway/heartbeat)
+    - [/gateway/troubleshooting#cron-and-heartbeat-delivery](/gateway/troubleshooting#cron-and-heartbeat-delivery)
+    - [/automation/cron-jobs#troubleshooting](/automation/cron-jobs#troubleshooting)
+    - [/gateway/heartbeat](/gateway/heartbeat)
 
     </Accordion>
 
@@ -338,7 +339,7 @@ flowchart TD
 
       - [/tools/exec](/tools/exec)
       - [/tools/exec-approvals](/tools/exec-approvals)
-      - [/gateway/security#runtime-expectation-drift](/gateway/security#runtime-expectation-drift)
+      - [/gateway/security#what-the-audit-checks-high-level](/gateway/security#what-the-audit-checks-high-level)
 
     </Accordion>
 
@@ -376,6 +377,7 @@ flowchart TD
       - [/tools/browser-wsl2-windows-remote-cdp-troubleshooting](/tools/browser-wsl2-windows-remote-cdp-troubleshooting)
 
     </Accordion>
+
   </AccordionGroup>
 
 ## Related

docs/plugins/codex-harness.md

@@ -0,0 +1,489 @@
+---
+title: "Codex Harness"
+summary: "Run OpenClaw embedded agent turns through the bundled Codex app-server harness"
+read_when:
+  - You want to use the bundled Codex app-server harness
+  - You need Codex model refs and config examples
+  - You want to disable PI fallback for Codex-only deployments
+---
+
+# Codex Harness
+
+The bundled `codex` plugin lets OpenClaw run embedded agent turns through the
+Codex app-server instead of the built-in PI harness.
+
+Use this when you want Codex to own the low-level agent session: model
+discovery, native thread resume, native compaction, and app-server execution.
+OpenClaw still owns chat channels, session files, model selection, tools,
+approvals, media delivery, and the visible transcript mirror.
+
+The harness is off by default. It is selected only when the `codex` plugin is
+enabled and the resolved model is a `codex/*` model, or when you explicitly
+force `embeddedHarness.runtime: "codex"` or `OPENCLAW_AGENT_RUNTIME=codex`.
+If you never configure `codex/*`, existing PI, OpenAI, Anthropic, Gemini, local,
+and custom-provider runs keep their current behavior.
+
+## Pick the right model prefix
+
+OpenClaw has separate routes for OpenAI and Codex-shaped access:
+
+| Model ref              | Runtime path                                 | Use when                                                                |
+| ---------------------- | -------------------------------------------- | ----------------------------------------------------------------------- |
+| `openai/gpt-5.4`       | OpenAI provider through OpenClaw/PI plumbing | You want direct OpenAI Platform API access with `OPENAI_API_KEY`.       |
+| `openai-codex/gpt-5.4` | OpenAI Codex OAuth provider through PI       | You want ChatGPT/Codex OAuth without the Codex app-server harness.      |
+| `codex/gpt-5.4`        | Bundled Codex provider plus Codex harness    | You want native Codex app-server execution for the embedded agent turn. |
+
+The Codex harness only claims `codex/*` model refs. Existing `openai/*`,
+`openai-codex/*`, Anthropic, Gemini, xAI, local, and custom provider refs keep
+their normal paths.
+
+## Requirements
+
+- OpenClaw with the bundled `codex` plugin available.
+- Codex app-server `0.118.0` or newer.
+- Codex auth available to the app-server process.
+
+The plugin blocks older or unversioned app-server handshakes. That keeps
+OpenClaw on the protocol surface it has been tested against.
+
+For live and Docker smoke tests, auth usually comes from `OPENAI_API_KEY`, plus
+optional Codex CLI files such as `~/.codex/auth.json` and
+`~/.codex/config.toml`. Use the same auth material your local Codex app-server
+uses.
+
+## Minimal config
+
+Use `codex/gpt-5.4`, enable the bundled plugin, and force the `codex` harness:
+
+```json5
+{
+  plugins: {
+    entries: {
+      codex: {
+        enabled: true,
+      },
+    },
+  },
+  agents: {
+    defaults: {
+      model: "codex/gpt-5.4",
+      embeddedHarness: {
+        runtime: "codex",
+        fallback: "none",
+      },
+    },
+  },
+}
+```
+
+If your config uses `plugins.allow`, include `codex` there too:
+
+```json5
+{
+  plugins: {
+    allow: ["codex"],
+    entries: {
+      codex: {
+        enabled: true,
+      },
+    },
+  },
+}
+```
+
+Setting `agents.defaults.model` or an agent model to `codex/<model>` also
+auto-enables the bundled `codex` plugin. The explicit plugin entry is still
+useful in shared configs because it makes the deployment intent obvious.
+
+## Add Codex without replacing other models
+
+Keep `runtime: "auto"` when you want Codex for `codex/*` models and PI for
+everything else:
+
+```json5
+{
+  plugins: {
+    entries: {
+      codex: {
+        enabled: true,
+      },
+    },
+  },
+  agents: {
+    defaults: {
+      model: {
+        primary: "codex/gpt-5.4",
+        fallbacks: ["openai/gpt-5.4", "anthropic/claude-opus-4-6"],
+      },
+      models: {
+        "codex/gpt-5.4": { alias: "codex" },
+        "codex/gpt-5.4-mini": { alias: "codex-mini" },
+        "openai/gpt-5.4": { alias: "gpt" },
+        "anthropic/claude-opus-4-6": { alias: "opus" },
+      },
+      embeddedHarness: {
+        runtime: "auto",
+        fallback: "pi",
+      },
+    },
+  },
+}
+```
+
+With this shape:
+
+- `/model codex` or `/model codex/gpt-5.4` uses the Codex app-server harness.
+- `/model gpt` or `/model openai/gpt-5.4` uses the OpenAI provider path.
+- `/model opus` uses the Anthropic provider path.
+- If a non-Codex model is selected, PI remains the compatibility harness.
+
+## Codex-only deployments
+
+Disable PI fallback when you need to prove that every embedded agent turn uses
+the Codex harness:
+
+```json5
+{
+  agents: {
+    defaults: {
+      model: "codex/gpt-5.4",
+      embeddedHarness: {
+        runtime: "codex",
+        fallback: "none",
+      },
+    },
+  },
+}
+```
+
+Environment override:
+
+```bash
+OPENCLAW_AGENT_RUNTIME=codex \
+OPENCLAW_AGENT_HARNESS_FALLBACK=none \
+openclaw gateway run
+```
+
+With fallback disabled, OpenClaw fails early if the Codex plugin is disabled,
+the requested model is not a `codex/*` ref, the app-server is too old, or the
+app-server cannot start.
+
+## Per-agent Codex
+
+You can make one agent Codex-only while the default agent keeps normal
+auto-selection:
+
+```json5
+{
+  agents: {
+    defaults: {
+      embeddedHarness: {
+        runtime: "auto",
+        fallback: "pi",
+      },
+    },
+    list: [
+      {
+        id: "main",
+        default: true,
+        model: "anthropic/claude-opus-4-6",
+      },
+      {
+        id: "codex",
+        name: "Codex",
+        model: "codex/gpt-5.4",
+        embeddedHarness: {
+          runtime: "codex",
+          fallback: "none",
+        },
+      },
+    ],
+  },
+}
+```
+
+Use normal session commands to switch agents and models. `/new` creates a fresh
+OpenClaw session and the Codex harness creates or resumes its sidecar app-server
+thread as needed. `/reset` clears the OpenClaw session binding for that thread.
+
+## Model discovery
+
+By default, the Codex plugin asks the app-server for available models. If
+discovery fails or times out, it uses the bundled fallback catalog:
+
+- `codex/gpt-5.4`
+- `codex/gpt-5.4-mini`
+- `codex/gpt-5.2`
+
+You can tune discovery under `plugins.entries.codex.config.discovery`:
+
+```json5
+{
+  plugins: {
+    entries: {
+      codex: {
+        enabled: true,
+        config: {
+          discovery: {
+            enabled: true,
+            timeoutMs: 2500,
+          },
+        },
+      },
+    },
+  },
+}
+```
+
+Disable discovery when you want startup to avoid probing Codex and stick to the
+fallback catalog:
+
+```json5
+{
+  plugins: {
+    entries: {
+      codex: {
+        enabled: true,
+        config: {
+          discovery: {
+            enabled: false,
+          },
+        },
+      },
+    },
+  },
+}
+```
+
+## App-server connection and policy
+
+By default, the plugin starts Codex locally with:
+
+```bash
+codex app-server --listen stdio://
+```
+
+You can keep that default and only tune Codex native policy:
+
+```json5
+{
+  plugins: {
+    entries: {
+      codex: {
+        enabled: true,
+        config: {
+          appServer: {
+            approvalPolicy: "on-request",
+            sandbox: "workspace-write",
+            serviceTier: "priority",
+          },
+        },
+      },
+    },
+  },
+}
+```
+
+For an already-running app-server, use WebSocket transport:
+
+```json5
+{
+  plugins: {
+    entries: {
+      codex: {
+        enabled: true,
+        config: {
+          appServer: {
+            transport: "websocket",
+            url: "ws://127.0.0.1:39175",
+            authToken: "${CODEX_APP_SERVER_TOKEN}",
+            requestTimeoutMs: 60000,
+          },
+        },
+      },
+    },
+  },
+}
+```
+
+Supported `appServer` fields:
+
+| Field               | Default                                  | Meaning                                                                  |
+| ------------------- | ---------------------------------------- | ------------------------------------------------------------------------ |
+| `transport`         | `"stdio"`                                | `"stdio"` spawns Codex; `"websocket"` connects to `url`.                 |
+| `command`           | `"codex"`                                | Executable for stdio transport.                                          |
+| `args`              | `["app-server", "--listen", "stdio://"]` | Arguments for stdio transport.                                           |
+| `url`               | unset                                    | WebSocket app-server URL.                                                |
+| `authToken`         | unset                                    | Bearer token for WebSocket transport.                                    |
+| `headers`           | `{}`                                     | Extra WebSocket headers.                                                 |
+| `requestTimeoutMs`  | `60000`                                  | Timeout for app-server control-plane calls.                              |
+| `approvalPolicy`    | `"never"`                                | Native Codex approval policy sent to thread start/resume/turn.           |
+| `sandbox`           | `"workspace-write"`                      | Native Codex sandbox mode sent to thread start/resume.                   |
+| `approvalsReviewer` | `"user"`                                 | Use `"guardian_subagent"` to let Codex guardian review native approvals. |
+| `serviceTier`       | unset                                    | Optional Codex service tier, for example `"priority"`.                   |
+
+The older environment variables still work as fallbacks for local testing when
+the matching config field is unset:
+
+- `OPENCLAW_CODEX_APP_SERVER_BIN`
+- `OPENCLAW_CODEX_APP_SERVER_ARGS`
+- `OPENCLAW_CODEX_APP_SERVER_APPROVAL_POLICY`
+- `OPENCLAW_CODEX_APP_SERVER_SANDBOX`
+- `OPENCLAW_CODEX_APP_SERVER_GUARDIAN=1`
+
+Config is preferred for repeatable deployments.
+
+## Common recipes
+
+Local Codex with default stdio transport:
+
+```json5
+{
+  plugins: {
+    entries: {
+      codex: {
+        enabled: true,
+      },
+    },
+  },
+}
+```
+
+Codex-only harness validation, with PI fallback disabled:
+
+```json5
+{
+  embeddedHarness: {
+    fallback: "none",
+  },
+  plugins: {
+    entries: {
+      codex: {
+        enabled: true,
+      },
+    },
+  },
+}
+```
+
+Guardian-reviewed Codex approvals:
+
+```json5
+{
+  plugins: {
+    entries: {
+      codex: {
+        enabled: true,
+        config: {
+          appServer: {
+            approvalPolicy: "on-request",
+            approvalsReviewer: "guardian_subagent",
+            sandbox: "workspace-write",
+          },
+        },
+      },
+    },
+  },
+}
+```
+
+Remote app-server with explicit headers:
+
+```json5
+{
+  plugins: {
+    entries: {
+      codex: {
+        enabled: true,
+        config: {
+          appServer: {
+            transport: "websocket",
+            url: "ws://gateway-host:39175",
+            headers: {
+              "X-OpenClaw-Agent": "main",
+            },
+          },
+        },
+      },
+    },
+  },
+}
+```
+
+Model switching stays OpenClaw-controlled. When an OpenClaw session is attached
+to an existing Codex thread, the next turn sends the currently selected
+`codex/*` model, provider, approval policy, sandbox, and service tier to
+app-server again. Switching from `codex/gpt-5.4` to `codex/gpt-5.2` keeps the
+thread binding but asks Codex to continue with the newly selected model.
+
+## Codex command
+
+The bundled plugin registers `/codex` as an authorized slash command. It is
+generic and works on any channel that supports OpenClaw text commands.
+
+Common forms:
+
+- `/codex status` shows live app-server connectivity, models, account, rate limits, MCP servers, and skills.
+- `/codex models` lists live Codex app-server models.
+- `/codex threads [filter]` lists recent Codex threads.
+- `/codex resume <thread-id>` attaches the current OpenClaw session to an existing Codex thread.
+- `/codex compact` asks Codex app-server to compact the attached thread.
+- `/codex review` starts Codex native review for the attached thread.
+- `/codex account` shows account and rate-limit status.
+- `/codex mcp` lists Codex app-server MCP server status.
+- `/codex skills` lists Codex app-server skills.
+
+`/codex resume` writes the same sidecar binding file that the harness uses for
+normal turns. On the next message, OpenClaw resumes that Codex thread, passes the
+currently selected OpenClaw `codex/*` model into app-server, and keeps extended
+history enabled.
+
+The command surface requires Codex app-server `0.118.0` or newer. Individual
+control methods are reported as `unsupported by this Codex app-server` if a
+future or custom app-server does not expose that JSON-RPC method.
+
+## Tools, media, and compaction
+
+The Codex harness changes the low-level embedded agent executor only.
+
+OpenClaw still builds the tool list and receives dynamic tool results from the
+harness. Text, images, video, music, TTS, approvals, and messaging-tool output
+continue through the normal OpenClaw delivery path.
+
+When the selected model uses the Codex harness, native thread compaction is
+delegated to Codex app-server. OpenClaw keeps a transcript mirror for channel
+history, search, `/new`, `/reset`, and future model or harness switching. The
+mirror includes the user prompt, final assistant text, and lightweight Codex
+reasoning or plan records when the app-server emits them.
+
+Media generation does not require PI. Image, video, music, PDF, TTS, and media
+understanding continue to use the matching provider/model settings such as
+`agents.defaults.imageGenerationModel`, `videoGenerationModel`, `pdfModel`, and
+`messages.tts`.
+
+## Troubleshooting
+
+**Codex does not appear in `/model`:** enable `plugins.entries.codex.enabled`,
+set a `codex/*` model ref, or check whether `plugins.allow` excludes `codex`.
+
+**OpenClaw falls back to PI:** set `embeddedHarness.fallback: "none"` or
+`OPENCLAW_AGENT_HARNESS_FALLBACK=none` while testing.
+
+**The app-server is rejected:** upgrade Codex so the app-server handshake
+reports version `0.118.0` or newer.
+
+**Model discovery is slow:** lower `plugins.entries.codex.config.discovery.timeoutMs`
+or disable discovery.
+
+**WebSocket transport fails immediately:** check `appServer.url`, `authToken`,
+and that the remote app-server speaks the same Codex app-server protocol version.
+
+**A non-Codex model uses PI:** that is expected. The Codex harness only claims
+`codex/*` model refs.
+
+## Related
+
+- [Agent Harness Plugins](/plugins/sdk-agent-harness)
+- [Model Providers](/concepts/model-providers)
+- [Configuration Reference](/gateway/configuration-reference)
+- [Testing](/help/testing#live-codex-app-server-harness-smoke)

docs/plugins/sdk-agent-harness.md

@@ -0,0 +1,264 @@
+---
+title: "Agent Harness Plugins"
+sidebarTitle: "Agent Harness"
+summary: "Experimental SDK surface for plugins that replace the low level embedded agent executor"
+read_when:
+  - You are changing the embedded agent runtime or harness registry
+  - You are registering an agent harness from a bundled or trusted plugin
+  - You need to understand how the Codex plugin relates to model providers
+---
+
+# Agent Harness Plugins
+
+An **agent harness** is the low level executor for one prepared OpenClaw agent
+turn. It is not a model provider, not a channel, and not a tool registry.
+
+Use this surface only for bundled or trusted native plugins. The contract is
+still experimental because the parameter types intentionally mirror the current
+embedded runner.
+
+## When to use a harness
+
+Register an agent harness when a model family has its own native session
+runtime and the normal OpenClaw provider transport is the wrong abstraction.
+
+Examples:
+
+- a native coding-agent server that owns threads and compaction
+- a local CLI or daemon that must stream native plan/reasoning/tool events
+- a model runtime that needs its own resume id in addition to the OpenClaw
+  session transcript
+
+Do **not** register a harness just to add a new LLM API. For normal HTTP or
+WebSocket model APIs, build a [provider plugin](/plugins/sdk-provider-plugins).
+
+## What core still owns
+
+Before a harness is selected, OpenClaw has already resolved:
+
+- provider and model
+- runtime auth state
+- thinking level and context budget
+- the OpenClaw transcript/session file
+- workspace, sandbox, and tool policy
+- channel reply callbacks and streaming callbacks
+- model fallback and live model switching policy
+
+That split is intentional. A harness runs a prepared attempt; it does not pick
+providers, replace channel delivery, or silently switch models.
+
+## Register a harness
+
+**Import:** `openclaw/plugin-sdk/agent-harness`
+
+```typescript
+import type { AgentHarness } from "openclaw/plugin-sdk/agent-harness";
+import { definePluginEntry } from "openclaw/plugin-sdk/plugin-entry";
+
+const myHarness: AgentHarness = {
+  id: "my-harness",
+  label: "My native agent harness",
+
+  supports(ctx) {
+    return ctx.provider === "my-provider"
+      ? { supported: true, priority: 100 }
+      : { supported: false };
+  },
+
+  async runAttempt(params) {
+    // Start or resume your native thread.
+    // Use params.prompt, params.tools, params.images, params.onPartialReply,
+    // params.onAgentEvent, and the other prepared attempt fields.
+    return await runMyNativeTurn(params);
+  },
+};
+
+export default definePluginEntry({
+  id: "my-native-agent",
+  name: "My Native Agent",
+  description: "Runs selected models through a native agent daemon.",
+  register(api) {
+    api.registerAgentHarness(myHarness);
+  },
+});
+```
+
+## Selection policy
+
+OpenClaw chooses a harness after provider/model resolution:
+
+1. `OPENCLAW_AGENT_RUNTIME=<id>` forces a registered harness with that id.
+2. `OPENCLAW_AGENT_RUNTIME=pi` forces the built-in PI harness.
+3. `OPENCLAW_AGENT_RUNTIME=auto` asks registered harnesses if they support the
+   resolved provider/model.
+4. If no registered harness matches, OpenClaw uses PI unless PI fallback is
+   disabled.
+
+Forced plugin harness failures surface as run failures. In `auto` mode,
+OpenClaw may fall back to PI when the selected plugin harness fails before a
+turn has produced side effects. Set `OPENCLAW_AGENT_HARNESS_FALLBACK=none` or
+`embeddedHarness.fallback: "none"` to make that fallback a hard failure instead.
+
+The bundled Codex plugin registers `codex` as its harness id. Core treats that
+as an ordinary plugin harness id; Codex-specific aliases belong in the plugin
+or operator config, not in the shared runtime selector.
+
+## Provider plus harness pairing
+
+Most harnesses should also register a provider. The provider makes model refs,
+auth status, model metadata, and `/model` selection visible to the rest of
+OpenClaw. The harness then claims that provider in `supports(...)`.
+
+The bundled Codex plugin follows this pattern:
+
+- provider id: `codex`
+- user model refs: `codex/gpt-5.4`, `codex/gpt-5.2`, or another model returned
+  by the Codex app server
+- harness id: `codex`
+- auth: synthetic provider availability, because the Codex harness owns the
+  native Codex login/session
+- app-server request: OpenClaw sends the bare model id to Codex and lets the
+  harness talk to the native app-server protocol
+
+The Codex plugin is additive. Plain `openai/gpt-*` refs remain OpenAI provider
+refs and continue to use the normal OpenClaw provider path. Select `codex/gpt-*`
+when you want Codex-managed auth, Codex model discovery, native threads, and
+Codex app-server execution. `/model` can switch among the Codex models returned
+by the Codex app server without requiring OpenAI provider credentials.
+
+For operator setup, model prefix examples, and Codex-only configs, see
+[Codex Harness](/plugins/codex-harness).
+
+OpenClaw requires Codex app-server `0.118.0` or newer. The Codex plugin checks
+the app-server initialize handshake and blocks older or unversioned servers so
+OpenClaw only runs against the protocol surface it has been tested with.
+
+## Disable PI fallback
+
+By default, OpenClaw runs embedded agents with `agents.defaults.embeddedHarness`
+set to `{ runtime: "auto", fallback: "pi" }`. In `auto` mode, registered plugin
+harnesses can claim a provider/model pair. If none match, or if an auto-selected
+plugin harness fails before producing output, OpenClaw falls back to PI.
+
+Set `fallback: "none"` when you need to prove that a plugin harness is the only
+runtime being exercised. This disables automatic PI fallback; it does not block
+an explicit `runtime: "pi"` or `OPENCLAW_AGENT_RUNTIME=pi`.
+
+For Codex-only embedded runs:
+
+```json
+{
+  "agents": {
+    "defaults": {
+      "model": "codex/gpt-5.4",
+      "embeddedHarness": {
+        "runtime": "codex",
+        "fallback": "none"
+      }
+    }
+  }
+}
+```
+
+If you want any registered plugin harness to claim matching models but never
+want OpenClaw to silently fall back to PI, keep `runtime: "auto"` and disable
+the fallback:
+
+```json
+{
+  "agents": {
+    "defaults": {
+      "embeddedHarness": {
+        "runtime": "auto",
+        "fallback": "none"
+      }
+    }
+  }
+}
+```
+
+Per-agent overrides use the same shape:
+
+```json
+{
+  "agents": {
+    "defaults": {
+      "embeddedHarness": {
+        "runtime": "auto",
+        "fallback": "pi"
+      }
+    },
+    "list": [
+      {
+        "id": "codex-only",
+        "model": "codex/gpt-5.4",
+        "embeddedHarness": {
+          "runtime": "codex",
+          "fallback": "none"
+        }
+      }
+    ]
+  }
+}
+```
+
+`OPENCLAW_AGENT_RUNTIME` still overrides the configured runtime. Use
+`OPENCLAW_AGENT_HARNESS_FALLBACK=none` to disable PI fallback from the
+environment.
+
+```bash
+OPENCLAW_AGENT_RUNTIME=codex \
+OPENCLAW_AGENT_HARNESS_FALLBACK=none \
+openclaw gateway run
+```
+
+With fallback disabled, a session fails early when the requested harness is not
+registered, does not support the resolved provider/model, or fails before
+producing turn side effects. That is intentional for Codex-only deployments and
+for live tests that must prove the Codex app-server path is actually in use.
+
+This setting only controls the embedded agent harness. It does not disable
+image, video, music, TTS, PDF, or other provider-specific model routing.
+
+## Native sessions and transcript mirror
+
+A harness may keep a native session id, thread id, or daemon-side resume token.
+Keep that binding explicitly associated with the OpenClaw session, and keep
+mirroring user-visible assistant/tool output into the OpenClaw transcript.
+
+The OpenClaw transcript remains the compatibility layer for:
+
+- channel-visible session history
+- transcript search and indexing
+- switching back to the built-in PI harness on a later turn
+- generic `/new`, `/reset`, and session deletion behavior
+
+If your harness stores a sidecar binding, implement `reset(...)` so OpenClaw can
+clear it when the owning OpenClaw session is reset.
+
+## Tool and media results
+
+Core constructs the OpenClaw tool list and passes it into the prepared attempt.
+When a harness executes a dynamic tool call, return the tool result back through
+the harness result shape instead of sending channel media yourself.
+
+This keeps text, image, video, music, TTS, approval, and messaging-tool outputs
+on the same delivery path as PI-backed runs.
+
+## Current limitations
+
+- The public import path is generic, but some attempt/result type aliases still
+  carry `Pi` names for compatibility.
+- Third-party harness installation is experimental. Prefer provider plugins
+  until you need a native session runtime.
+- Harness switching is supported across turns. Do not switch harnesses in the
+  middle of a turn after native tools, approvals, assistant text, or message
+  sends have started.
+
+## Related
+
+- [SDK Overview](/plugins/sdk-overview)
+- [Runtime Helpers](/plugins/sdk-runtime)
+- [Provider Plugins](/plugins/sdk-provider-plugins)
+- [Codex Harness](/plugins/codex-harness)
+- [Model Providers](/concepts/model-providers)

docs/plugins/sdk-channel-plugins.md

@@ -256,7 +256,7 @@ should use `resolveInboundMentionDecision({ facts, policy })`.
   <Step title="Package and manifest">
     Create the standard plugin files. The `channel` field in `package.json` is
     what makes this a channel plugin. For the full package-metadata surface,
-    see [Plugin Setup and Config](/plugins/sdk-setup#openclawchannel):
+    see [Plugin Setup and Config](/plugins/sdk-setup#openclaw-channel):
 
     <CodeGroup>
     ```json package.json

docs/plugins/sdk-overview.md

@@ -219,6 +219,7 @@ explicitly promotes one as public.
     | `plugin-sdk/models-provider-runtime` | `/models` command/provider reply helpers |
     | `plugin-sdk/skill-commands-runtime` | Skill command listing helpers |
     | `plugin-sdk/native-command-registry` | Native command registry/build/serialize helpers |
+    | `plugin-sdk/agent-harness` | Experimental trusted-plugin surface for low-level agent harnesses: harness types, active-run steer/abort helpers, OpenClaw tool bridge helpers, and attempt result utilities |
     | `plugin-sdk/provider-zai-endpoint` | Z.AI endpoint detection helpers |
     | `plugin-sdk/infra-runtime` | System event/heartbeat helpers |
     | `plugin-sdk/collection-runtime` | Small bounded cache helpers |
@@ -302,20 +303,21 @@ methods:
 
 ### Capability registration
 
-| Method                                           | What it registers                |
-| ------------------------------------------------ | -------------------------------- |
-| `api.registerProvider(...)`                      | Text inference (LLM)             |
-| `api.registerCliBackend(...)`                    | Local CLI inference backend      |
-| `api.registerChannel(...)`                       | Messaging channel                |
-| `api.registerSpeechProvider(...)`                | Text-to-speech / STT synthesis   |
-| `api.registerRealtimeTranscriptionProvider(...)` | Streaming realtime transcription |
-| `api.registerRealtimeVoiceProvider(...)`         | Duplex realtime voice sessions   |
-| `api.registerMediaUnderstandingProvider(...)`    | Image/audio/video analysis       |
-| `api.registerImageGenerationProvider(...)`       | Image generation                 |
-| `api.registerMusicGenerationProvider(...)`       | Music generation                 |
-| `api.registerVideoGenerationProvider(...)`       | Video generation                 |
-| `api.registerWebFetchProvider(...)`              | Web fetch / scrape provider      |
-| `api.registerWebSearchProvider(...)`             | Web search                       |
+| Method                                           | What it registers                     |
+| ------------------------------------------------ | ------------------------------------- |
+| `api.registerProvider(...)`                      | Text inference (LLM)                  |
+| `api.registerAgentHarness(...)`                  | Experimental low-level agent executor |
+| `api.registerCliBackend(...)`                    | Local CLI inference backend           |
+| `api.registerChannel(...)`                       | Messaging channel                     |
+| `api.registerSpeechProvider(...)`                | Text-to-speech / STT synthesis        |
+| `api.registerRealtimeTranscriptionProvider(...)` | Streaming realtime transcription      |
+| `api.registerRealtimeVoiceProvider(...)`         | Duplex realtime voice sessions        |
+| `api.registerMediaUnderstandingProvider(...)`    | Image/audio/video analysis            |
+| `api.registerImageGenerationProvider(...)`       | Image generation                      |
+| `api.registerMusicGenerationProvider(...)`       | Music generation                      |
+| `api.registerVideoGenerationProvider(...)`       | Video generation                      |
+| `api.registerWebFetchProvider(...)`              | Web fetch / scrape provider           |
+| `api.registerWebSearchProvider(...)`             | Web search                            |
 
 ### Tools and commands
 

docs/plugins/sdk-provider-plugins.md

@@ -20,6 +20,13 @@ API key auth, and dynamic model resolution.
   structure and manifest setup.
 </Info>
 
+<Tip>
+  Provider plugins add models to OpenClaw's normal inference loop. If the model
+  must run through a native agent daemon that owns threads, compaction, or tool
+  events, pair the provider with an [agent harness](/plugins/sdk-agent-harness)
+  instead of putting daemon protocol details in core.
+</Tip>
+
 ## Walkthrough
 
 <Steps>
@@ -168,6 +175,28 @@ API key auth, and dynamic model resolution.
     `openclaw onboard --acme-ai-api-key <key>` and select
     `acme-ai/acme-large` as their model.
 
+    If the upstream provider uses different control tokens than OpenClaw, add a
+    small bidirectional text transform instead of replacing the stream path:
+
+    ```typescript
+    api.registerTextTransforms({
+      input: [
+        { from: /red basket/g, to: "blue basket" },
+        { from: /paper ticket/g, to: "digital ticket" },
+        { from: /left shelf/g, to: "right shelf" },
+      ],
+      output: [
+        { from: /blue basket/g, to: "red basket" },
+        { from: /digital ticket/g, to: "paper ticket" },
+        { from: /right shelf/g, to: "left shelf" },
+      ],
+    });
+    ```
+
+    `input` rewrites the final system prompt and text message content before
+    transport. `output` rewrites assistant text deltas and final text before
+    OpenClaw parses its own control markers or channel delivery.
+
     For bundled providers that only register one text provider with API-key
     auth plus a single catalog-backed runtime, prefer the narrower
     `defineSingleProviderPluginEntry(...)` helper:

docs/plugins/sdk-runtime.md

@@ -50,9 +50,9 @@ const timeoutMs = api.runtime.agent.resolveAgentTimeoutMs(cfg);
 // Ensure workspace exists
 await api.runtime.agent.ensureAgentWorkspace(cfg);
 
-// Run an embedded Pi agent
+// Run an embedded agent turn
 const agentDir = api.runtime.agent.resolveAgentDir(cfg);
-const result = await api.runtime.agent.runEmbeddedPiAgent({
+const result = await api.runtime.agent.runEmbeddedAgent({
   sessionId: "my-plugin:task-1",
   runId: crypto.randomUUID(),
   sessionFile: path.join(agentDir, "sessions", "my-plugin-task-1.jsonl"),
@@ -62,6 +62,12 @@ const result = await api.runtime.agent.runEmbeddedPiAgent({
 });
 ```
 
+`runEmbeddedAgent(...)` is the neutral helper for starting a normal OpenClaw
+agent turn from plugin code. It uses the same provider/model resolution and
+agent-harness selection as channel-triggered replies.
+
+`runEmbeddedPiAgent(...)` remains as a compatibility alias.
+
 **Session store helpers** are under `api.runtime.agent.session`:
 
 ```typescript

docs/providers/fal.md

@@ -69,15 +69,36 @@ The bundled `fal` video-generation provider defaults to
 
 - Modes: text-to-video and single-image reference flows
 - Runtime: queue-backed submit/status/result flow for long-running jobs
+- HeyGen video-agent model ref:
+  - `fal/fal-ai/heygen/v2/video-agent`
+- Seedance 2.0 model refs:
+  - `fal/bytedance/seedance-2.0/fast/text-to-video`
+  - `fal/bytedance/seedance-2.0/fast/image-to-video`
+  - `fal/bytedance/seedance-2.0/text-to-video`
+  - `fal/bytedance/seedance-2.0/image-to-video`
 
-To use fal as the default video provider:
+To use Seedance 2.0 as the default video model:
 
 ```json5
 {
   agents: {
     defaults: {
       videoGenerationModel: {
-        primary: "fal/fal-ai/minimax/video-01-live",
+        primary: "fal/bytedance/seedance-2.0/fast/text-to-video",
+      },
+    },
+  },
+}
+```
+
+To use HeyGen video-agent as the default video model:
+
+```json5
+{
+  agents: {
+    defaults: {
+      videoGenerationModel: {
+        primary: "fal/fal-ai/heygen/v2/video-agent",
       },
     },
   },

docs/reference/RELEASING.md

@@ -69,9 +69,9 @@ OpenClaw has three public release lanes:
 - npm release preflight fails closed unless the tarball includes both
   `dist/control-ui/index.html` and a non-empty `dist/control-ui/assets/` payload
   so we do not ship an empty browser dashboard again
-- If the release work touched CI planning, extension timing manifests, or fast
-  test matrices, regenerate and review the planner-owned `checks-fast-extensions`
-  workflow matrix outputs from `.github/workflows/ci.yml`
+- If the release work touched CI planning, extension timing manifests, or
+  extension test matrices, regenerate and review the planner-owned
+  `checks-node-extensions` workflow matrix outputs from `.github/workflows/ci.yml`
   before approval so release notes do not describe a stale CI layout
 - Stable macOS release readiness also includes the updater surfaces:
   - the GitHub release must end up with the packaged `.zip`, `.dmg`, and `.dSYM.zip`

docs/tools/browser.md

@@ -146,7 +146,7 @@ Browser settings live in `~/.openclaw/openclaw.json`.
   browser: {
     enabled: true, // default: true
     ssrfPolicy: {
-      dangerouslyAllowPrivateNetwork: true, // default trusted-network mode
+      // dangerouslyAllowPrivateNetwork: true, // opt in only for trusted private-network access
       // allowPrivateNetwork: true, // legacy alias
       // hostnameAllowlist: ["*.example.com", "example.com"],
       // allowedHostnames: ["localhost"],
@@ -191,7 +191,7 @@ Notes:
 - `remoteCdpHandshakeTimeoutMs` applies to remote CDP WebSocket reachability checks.
 - Browser navigation/open-tab is SSRF-guarded before navigation and best-effort re-checked on final `http(s)` URL after navigation.
 - In strict SSRF mode, remote CDP endpoint discovery/probes (`cdpUrl`, including `/json/version` lookups) are checked too.
-- `browser.ssrfPolicy.dangerouslyAllowPrivateNetwork` defaults to `true` (trusted-network model). Set it to `false` for strict public-only browsing.
+- `browser.ssrfPolicy.dangerouslyAllowPrivateNetwork` is disabled by default. Set it to `true` only when you intentionally trust private-network browser access.
 - `browser.ssrfPolicy.allowPrivateNetwork` remains supported as a legacy alias for compatibility.
 - `attachOnly: true` means “never launch a local browser; only attach if it is already running.”
 - `color` + per-profile `color` tint the browser UI so you can see which profile is active.

docs/tools/reactions.md

@@ -68,7 +68,7 @@ tool with the `react` action. Reaction behavior varies by channel.
 Per-channel `reactionLevel` config controls how broadly the agent uses reactions. Values are typically `off`, `ack`, `minimal`, or `extensive`.
 
 - [Telegram reactionLevel](/channels/telegram#reaction-notifications) — `channels.telegram.reactionLevel`
-- [WhatsApp reactionLevel](/channels/whatsapp#reactions) — `channels.whatsapp.reactionLevel`
+- [WhatsApp reactionLevel](/channels/whatsapp#reaction-level) — `channels.whatsapp.reactionLevel`
 
 Set `reactionLevel` on individual channels to tune how actively the agent reacts to messages on each platform.
 

docs/tools/slash-commands.md

@@ -152,6 +152,7 @@ Bundled plugins can add more slash commands. Current bundled commands in this re
 - `/phone status|arm <camera|screen|writes|all> [duration]|disarm` temporarily arms high-risk phone node commands.
 - `/voice status|list [limit]|set <voiceId|name>` manages Talk voice config. On Discord, the native command name is `/talkvoice`.
 - `/card ...` sends LINE rich card presets. See [LINE](/channels/line).
+- `/codex status|models|threads|resume|compact|review|account|mcp|skills` inspects and controls the bundled Codex app-server harness. See [Codex Harness](/plugins/codex-harness).
 - QQBot-only commands:
   - `/bot-ping`
   - `/bot-version`

docs/tools/video-generation.md

@@ -201,22 +201,50 @@ entries.
 }
 ```
 
+HeyGen video-agent on fal can be pinned with:
+
+```json5
+{
+  agents: {
+    defaults: {
+      videoGenerationModel: {
+        primary: "fal/fal-ai/heygen/v2/video-agent",
+      },
+    },
+  },
+}
+```
+
+Seedance 2.0 on fal can be pinned with:
+
+```json5
+{
+  agents: {
+    defaults: {
+      videoGenerationModel: {
+        primary: "fal/bytedance/seedance-2.0/fast/text-to-video",
+      },
+    },
+  },
+}
+```
+
 ## Provider notes
 
-| Provider | Notes                                                                                                                                                       |
-| -------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| Alibaba  | Uses DashScope/Model Studio async endpoint. Reference images and videos must be remote `http(s)` URLs.                                                      |
-| BytePlus | Single image reference only.                                                                                                                                |
-| ComfyUI  | Workflow-driven local or cloud execution. Supports text-to-video and image-to-video through the configured graph.                                           |
-| fal      | Uses queue-backed flow for long-running jobs. Single image reference only.                                                                                  |
-| Google   | Uses Gemini/Veo. Supports one image or one video reference.                                                                                                 |
-| MiniMax  | Single image reference only.                                                                                                                                |
-| OpenAI   | Only `size` override is forwarded. Other style overrides (`aspectRatio`, `resolution`, `audio`, `watermark`) are ignored with a warning.                    |
-| Qwen     | Same DashScope backend as Alibaba. Reference inputs must be remote `http(s)` URLs; local files are rejected upfront.                                        |
-| Runway   | Supports local files via data URIs. Video-to-video requires `runway/gen4_aleph`. Text-only runs expose `16:9` and `9:16` aspect ratios.                     |
-| Together | Single image reference only.                                                                                                                                |
-| Vydra    | Uses `https://www.vydra.ai/api/v1` directly to avoid auth-dropping redirects. `veo3` is bundled as text-to-video only; `kling` requires a remote image URL. |
-| xAI      | Supports text-to-video, image-to-video, and remote video edit/extend flows.                                                                                 |
+| Provider | Notes                                                                                                                                                                |
+| -------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| Alibaba  | Uses DashScope/Model Studio async endpoint. Reference images and videos must be remote `http(s)` URLs.                                                               |
+| BytePlus | Single image reference only.                                                                                                                                         |
+| ComfyUI  | Workflow-driven local or cloud execution. Supports text-to-video and image-to-video through the configured graph.                                                    |
+| fal      | Uses queue-backed flow for long-running jobs. Single image reference only. Includes HeyGen video-agent and Seedance 2.0 text-to-video and image-to-video model refs. |
+| Google   | Uses Gemini/Veo. Supports one image or one video reference.                                                                                                          |
+| MiniMax  | Single image reference only.                                                                                                                                         |
+| OpenAI   | Only `size` override is forwarded. Other style overrides (`aspectRatio`, `resolution`, `audio`, `watermark`) are ignored with a warning.                             |
+| Qwen     | Same DashScope backend as Alibaba. Reference inputs must be remote `http(s)` URLs; local files are rejected upfront.                                                 |
+| Runway   | Supports local files via data URIs. Video-to-video requires `runway/gen4_aleph`. Text-only runs expose `16:9` and `9:16` aspect ratios.                              |
+| Together | Single image reference only.                                                                                                                                         |
+| Vydra    | Uses `https://www.vydra.ai/api/v1` directly to avoid auth-dropping redirects. `veo3` is bundled as text-to-video only; `kling` requires a remote image URL.          |
+| xAI      | Supports text-to-video, image-to-video, and remote video edit/extend flows.                                                                                          |
 
 ## Provider capability modes
 

extensions/acpx/package.json

@@ -4,7 +4,7 @@
   "description": "OpenClaw ACP runtime backend",
   "type": "module",
   "dependencies": {
-    "acpx": "0.5.2"
+    "acpx": "0.5.3"
   },
   "devDependencies": {
     "@openclaw/plugin-sdk": "workspace:*"

extensions/acpx/src/runtime-internals/mcp-command-line.test.ts

@@ -2,7 +2,7 @@ import { describe, expect, it } from "vitest";
 
 type SplitCommandLine = (
   value: string,
-  platform?: NodeJS.Platform | string,
+  platform?: string,
 ) => {
   command: string;
   args: string[];

extensions/acpx/src/runtime.test.ts

@@ -1,11 +1,15 @@
-import type { AcpSessionStore } from "acpx/runtime";
 import { beforeEach, describe, expect, it, vi } from "vitest";
 import type { AcpRuntime } from "../runtime-api.js";
 import { AcpxRuntime } from "./runtime.js";
 
-function makeRuntime(baseStore: AcpSessionStore): {
+type TestSessionStore = {
+  load(sessionId: string): Promise<Record<string, unknown> | undefined>;
+  save(record: Record<string, unknown>): Promise<void>;
+};
+
+function makeRuntime(baseStore: TestSessionStore): {
   runtime: AcpxRuntime;
-  wrappedStore: AcpSessionStore & { markFresh: (sessionKey: string) => void };
+  wrappedStore: TestSessionStore & { markFresh: (sessionKey: string) => void };
   delegate: { close: AcpRuntime["close"] };
 } {
   const runtime = new AcpxRuntime({
@@ -22,7 +26,7 @@ function makeRuntime(baseStore: AcpSessionStore): {
     runtime,
     wrappedStore: (
       runtime as unknown as {
-        sessionStore: AcpSessionStore & { markFresh: (sessionKey: string) => void };
+        sessionStore: TestSessionStore & { markFresh: (sessionKey: string) => void };
       }
     ).sessionStore,
     delegate: (runtime as unknown as { delegate: { close: AcpRuntime["close"] } }).delegate,
@@ -35,7 +39,7 @@ describe("AcpxRuntime fresh reset wrapper", () => {
   });
 
   it("keeps stale persistent loads hidden until a fresh record is saved", async () => {
-    const baseStore: AcpSessionStore = {
+    const baseStore: TestSessionStore = {
       load: vi.fn(async () => ({ acpxRecordId: "stale" }) as never),
       save: vi.fn(async () => {}),
     };
@@ -68,7 +72,7 @@ describe("AcpxRuntime fresh reset wrapper", () => {
   });
 
   it("marks the session fresh after discardPersistentState close", async () => {
-    const baseStore: AcpSessionStore = {
+    const baseStore: TestSessionStore = {
       load: vi.fn(async () => ({ acpxRecordId: "stale" }) as never),
       save: vi.fn(async () => {}),
     };

extensions/acpx/src/service.test.ts

@@ -19,7 +19,7 @@ vi.mock("../runtime-api.js", () => ({
 
 vi.mock("./runtime.js", () => ({
   ACPX_BACKEND_ID: "acpx",
-  AcpxRuntime: class {},
+  AcpxRuntime: function AcpxRuntime() {},
   createAgentRegistry: vi.fn(() => ({})),
   createFileSessionStore: vi.fn(() => ({})),
 }));

extensions/active-memory/index.test.ts

@@ -111,7 +111,7 @@ describe("active-memory plugin", () => {
     runEmbeddedPiAgent.mockResolvedValue({
       payloads: [{ text: "- lemon pepper wings\n- blue cheese" }],
     });
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
   });
 
   afterEach(async () => {
@@ -406,7 +406,7 @@ describe("active-memory plugin", () => {
       agents: ["main"],
       allowedChatTypes: ["direct", "group"],
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
 
     const result = await hooks.before_prompt_build(
       { prompt: "what wings should we order?", messages: [] },
@@ -509,7 +509,7 @@ describe("active-memory plugin", () => {
       agents: ["main"],
       queryMode: "message",
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
 
     await hooks.before_prompt_build(
       {
@@ -537,7 +537,7 @@ describe("active-memory plugin", () => {
       queryMode: "message",
       promptStyle: "preference-only",
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
 
     await hooks.before_prompt_build(
       {
@@ -582,7 +582,7 @@ describe("active-memory plugin", () => {
       agents: ["main"],
       thinking: "medium",
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
 
     await hooks.before_prompt_build(
       {
@@ -608,7 +608,7 @@ describe("active-memory plugin", () => {
       agents: ["main"],
       promptAppend: "Prefer stable long-term preferences over one-off events.",
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
 
     await hooks.before_prompt_build(
       {
@@ -637,7 +637,7 @@ describe("active-memory plugin", () => {
       promptOverride: "Custom memory prompt. Return NONE or one user fact.",
       promptAppend: "Extra custom instruction.",
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
 
     await hooks.before_prompt_build(
       {
@@ -710,7 +710,7 @@ describe("active-memory plugin", () => {
     api.pluginConfig = {
       agents: ["main"],
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
 
     await hooks.before_prompt_build(
       { prompt: "what wings should i order? temp transcript", messages: [] },
@@ -735,7 +735,7 @@ describe("active-memory plugin", () => {
       agents: ["main"],
       modelFallbackPolicy: "resolved-only",
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
 
     const result = await hooks.before_prompt_build(
       { prompt: "what wings should i order? no fallback", messages: [] },
@@ -872,7 +872,7 @@ describe("active-memory plugin", () => {
       timeoutMs: 250,
       logging: true,
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
     let lastAbortSignal: AbortSignal | undefined;
     runEmbeddedPiAgent.mockImplementation(async (params: { abortSignal?: AbortSignal }) => {
       lastAbortSignal = params.abortSignal;
@@ -918,7 +918,7 @@ describe("active-memory plugin", () => {
       agents: ["main"],
       logging: true,
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
 
     await hooks.before_prompt_build(
       { prompt: "what wings should i order? session id cache", messages: [] },
@@ -1037,7 +1037,7 @@ describe("active-memory plugin", () => {
       agents: ["main"],
       queryMode: "message",
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
 
     await hooks.before_prompt_build(
       {
@@ -1065,7 +1065,7 @@ describe("active-memory plugin", () => {
       agents: ["main"],
       queryMode: "full",
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
 
     await hooks.before_prompt_build(
       {
@@ -1096,7 +1096,7 @@ describe("active-memory plugin", () => {
       agents: ["main"],
       queryMode: "recent",
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
 
     await hooks.before_prompt_build(
       {
@@ -1174,7 +1174,7 @@ describe("active-memory plugin", () => {
       agents: ["main"],
       maxSummaryChars: 40,
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
     runEmbeddedPiAgent.mockResolvedValueOnce({
       payloads: [
         {
@@ -1211,7 +1211,7 @@ describe("active-memory plugin", () => {
       agents: ["main"],
       maxSummaryChars: 90,
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
 
     await hooks.before_prompt_build(
       { prompt: "what wings should i order? prompt-count-check", messages: [] },
@@ -1261,7 +1261,7 @@ describe("active-memory plugin", () => {
       transcriptDir: "active-memory-subagents",
       logging: true,
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
     const mkdirSpy = vi.spyOn(fs, "mkdir").mockResolvedValue(undefined);
     const mkdtempSpy = vi.spyOn(fs, "mkdtemp");
     const rmSpy = vi.spyOn(fs, "rm").mockResolvedValue(undefined);
@@ -1305,7 +1305,7 @@ describe("active-memory plugin", () => {
       transcriptDir: "C:/temp/escape",
       logging: true,
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
     const mkdirSpy = vi.spyOn(fs, "mkdir").mockResolvedValue(undefined);
 
     await hooks.before_prompt_build(
@@ -1342,7 +1342,7 @@ describe("active-memory plugin", () => {
       transcriptDir: "active-memory-subagents",
       logging: true,
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
     const mkdirSpy = vi.spyOn(fs, "mkdir").mockResolvedValue(undefined);
 
     await hooks.before_prompt_build(
@@ -1409,7 +1409,7 @@ describe("active-memory plugin", () => {
       agents: ["main"],
       logging: true,
     };
-    plugin.register(api as unknown as OpenClawPluginApi);
+    await plugin.register(api as unknown as OpenClawPluginApi);
 
     for (let index = 0; index <= 1000; index += 1) {
       await hooks.before_prompt_build(

extensions/active-memory/index.ts

@@ -495,7 +495,7 @@ function normalizePluginConfig(pluginConfig: unknown): ResolvedActiveRecallPlugi
   return {
     enabled: raw.enabled !== false,
     agents: Array.isArray(raw.agents)
-      ? raw.agents.map((agentId) => String(agentId).trim()).filter(Boolean)
+      ? raw.agents.map((agentId) => agentId.trim()).filter(Boolean)
       : [],
     model: typeof raw.model === "string" && raw.model.trim() ? raw.model.trim() : undefined,
     modelFallbackPolicy:
@@ -1496,7 +1496,7 @@ export default definePluginEntry({
           agentId: effectiveAgentId,
           sessionKey: resolvedSessionKey,
         });
-        return;
+        return undefined;
       }
       if (!isEnabledForAgent(config, effectiveAgentId)) {
         await persistPluginStatusLines({
@@ -1504,7 +1504,7 @@ export default definePluginEntry({
           agentId: effectiveAgentId,
           sessionKey: resolvedSessionKey,
         });
-        return;
+        return undefined;
       }
       if (!isEligibleInteractiveSession(ctx)) {
         await persistPluginStatusLines({
@@ -1512,7 +1512,7 @@ export default definePluginEntry({
           agentId: effectiveAgentId,
           sessionKey: resolvedSessionKey,
         });
-        return;
+        return undefined;
       }
       if (
         !isAllowedChatType(config, {
@@ -1526,7 +1526,7 @@ export default definePluginEntry({
           agentId: effectiveAgentId,
           sessionKey: resolvedSessionKey,
         });
-        return;
+        return undefined;
       }
       const query = buildQuery({
         latestUserMessage: event.prompt,
@@ -1544,11 +1544,11 @@ export default definePluginEntry({
         currentModelId: ctx.modelId,
       });
       if (!result.summary) {
-        return;
+        return undefined;
       }
       const metadata = buildMetadata(result.summary);
       if (!metadata) {
-        return;
+        return undefined;
       }
       return {
         prependSystemContext: ACTIVE_MEMORY_PLUGIN_GUIDANCE,

extensions/amazon-bedrock/package.json

@@ -5,7 +5,7 @@
   "description": "OpenClaw Amazon Bedrock provider plugin",
   "type": "module",
   "dependencies": {
-    "@aws-sdk/client-bedrock": "3.1024.0"
+    "@aws-sdk/client-bedrock": "3.1028.0"
   },
   "devDependencies": {
     "@openclaw/plugin-sdk": "workspace:*"

extensions/anthropic/stream-wrappers.test.ts

@@ -18,7 +18,7 @@ function runWrapper(apiKey: string | undefined): Record<string, string> | undefi
     return {} as never;
   };
   const wrapper = createAnthropicBetaHeadersWrapper(base, [CONTEXT_1M_BETA]);
-  wrapper(
+  void wrapper(
     { provider: "anthropic", id: "claude-opus-4-6" } as never,
     {} as never,
     { apiKey } as never,
@@ -64,7 +64,7 @@ describe("anthropic stream wrappers", () => {
       extraParams: { context1m: true, serviceTier: "auto" },
     } as never);
 
-    wrapped?.(
+    void wrapped?.(
       { provider: "anthropic", api: "anthropic-messages", id: "claude-sonnet-4-6" } as never,
       {} as never,
       { apiKey: "sk-ant-oat01-oauth-token" } as never,
@@ -91,7 +91,7 @@ describe("anthropic stream wrappers", () => {
       extraParams: { context1m: true, serviceTier: "auto" },
     } as never);
 
-    wrapped?.(
+    void wrapped?.(
       { provider: "anthropic", api: "anthropic-messages", id: "claude-sonnet-4-6" } as never,
       {} as never,
       { apiKey: "sk-ant-api-123" } as never,
@@ -121,7 +121,7 @@ describe("createAnthropicFastModeWrapper", () => {
     };
 
     const wrapper = createAnthropicFastModeWrapper(base, params.enabled ?? true);
-    wrapper(
+    void wrapper(
       {
         provider: params.provider ?? "anthropic",
         api: params.api ?? "anthropic-messages",
@@ -177,7 +177,7 @@ describe("createAnthropicServiceTierWrapper", () => {
     };
 
     const wrapper = createAnthropicServiceTierWrapper(base, params.serviceTier ?? "auto");
-    wrapper(
+    void wrapper(
       {
         provider: params.provider ?? "anthropic",
         api: params.api ?? "anthropic-messages",

extensions/arcee/provider-catalog.ts

@@ -4,9 +4,7 @@ import { buildArceeModelDefinition, ARCEE_BASE_URL, ARCEE_MODEL_CATALOG } from "
 export const OPENROUTER_BASE_URL = "https://openrouter.ai/api/v1";
 
 function normalizeBaseUrl(baseUrl: string | undefined): string {
-  return String(baseUrl ?? "")
-    .trim()
-    .replace(/\/+$/, "");
+  return (baseUrl ?? "").trim().replace(/\/+$/, "");
 }
 
 export function isArceeOpenRouterBaseUrl(baseUrl: string | undefined): boolean {

extensions/bluebubbles/src/conversation-bindings.test.ts

@@ -0,0 +1,64 @@
+import type { OpenClawConfig } from "openclaw/plugin-sdk/config-runtime";
+import {
+  __testing as sessionBindingTesting,
+  getSessionBindingService,
+} from "openclaw/plugin-sdk/conversation-runtime";
+import { beforeEach, describe, expect, it } from "vitest";
+import { __testing, createBlueBubblesConversationBindingManager } from "./conversation-bindings.js";
+
+const baseCfg = {
+  session: { mainKey: "main", scope: "per-sender" },
+} satisfies OpenClawConfig;
+
+describe("BlueBubbles conversation bindings", () => {
+  beforeEach(() => {
+    sessionBindingTesting.resetSessionBindingAdaptersForTests();
+    __testing.resetBlueBubblesConversationBindingsForTests();
+  });
+
+  it("preserves existing metadata when rebinding the same conversation", async () => {
+    const manager = createBlueBubblesConversationBindingManager({
+      cfg: baseCfg,
+      accountId: "default",
+    });
+
+    manager.bindConversation({
+      conversationId: "chat-guid-1",
+      targetKind: "subagent",
+      targetSessionKey: "agent:main:subagent:child",
+      metadata: {
+        agentId: "codex",
+        label: "child",
+        boundBy: "system",
+      },
+    });
+
+    await getSessionBindingService().bind({
+      targetSessionKey: "agent:main:subagent:child",
+      targetKind: "subagent",
+      conversation: {
+        channel: "bluebubbles",
+        accountId: "default",
+        conversationId: "chat-guid-1",
+      },
+      placement: "current",
+      metadata: {
+        label: "child",
+      },
+    });
+
+    expect(
+      getSessionBindingService().resolveByConversation({
+        channel: "bluebubbles",
+        accountId: "default",
+        conversationId: "chat-guid-1",
+      }),
+    ).toMatchObject({
+      metadata: expect.objectContaining({
+        agentId: "codex",
+        label: "child",
+        boundBy: "system",
+      }),
+    });
+  });
+});

extensions/bluebubbles/src/conversation-route.test.ts

@@ -3,7 +3,7 @@ import {
   __testing as sessionBindingTesting,
   registerSessionBindingAdapter,
 } from "openclaw/plugin-sdk/conversation-runtime";
-import { beforeEach, describe, expect, it, vi } from "vitest";
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
 import { resolveBlueBubblesConversationRoute } from "./conversation-route.js";
 
 const baseCfg = {
@@ -18,6 +18,10 @@ describe("resolveBlueBubblesConversationRoute", () => {
     sessionBindingTesting.resetSessionBindingAdaptersForTests();
   });
 
+  afterEach(() => {
+    sessionBindingTesting.resetSessionBindingAdaptersForTests();
+  });
+
   it("lets runtime BlueBubbles conversation bindings override default routing", () => {
     const touch = vi.fn();
     registerSessionBindingAdapter({

extensions/bluebubbles/src/monitor-debounce.ts

@@ -89,9 +89,7 @@ function combineDebounceEntries(entries: BlueBubblesDebounceEntry[]): Normalized
   const latestTimestamp = timestamps.length > 0 ? Math.max(...timestamps) : first.timestamp;
 
   // Collect all message IDs for reference
-  const messageIds = entries
-    .map((e) => e.message.messageId)
-    .filter((id): id is string => Boolean(id));
+  const messageId = entries.map((e) => e.message.messageId).find((id): id is string => Boolean(id));
 
   // Prefer reply context from any entry that has it
   const entryWithReply = entries.find((e) => e.message.replyToId);
@@ -102,7 +100,7 @@ function combineDebounceEntries(entries: BlueBubblesDebounceEntry[]): Normalized
     attachments: allAttachments.length > 0 ? allAttachments : first.attachments,
     timestamp: latestTimestamp,
     // Use first message's ID as primary (for reply reference), but we've coalesced others
-    messageId: messageIds[0] ?? first.messageId,
+    messageId: messageId ?? first.messageId,
     // Preserve reply context if present
     replyToId: entryWithReply?.message.replyToId ?? first.replyToId,
     replyToBody: entryWithReply?.message.replyToBody ?? first.replyToBody,

extensions/bluebubbles/src/monitor-processing.ts

@@ -1169,7 +1169,7 @@ export async function processMessage(
         isDirect: !isGroup,
         isGroup,
         isMentionableGroup: isGroup,
-        requireMention: Boolean(requireMention),
+        requireMention,
         canDetectMention,
         effectiveWasMentioned,
         shouldBypassMention,

extensions/bluebubbles/src/monitor.test.ts

@@ -1370,7 +1370,7 @@ describe("BlueBubbles webhook monitor", () => {
       mockDispatchReplyWithBufferedBlockDispatcher.mockImplementationOnce(async (params) => {
         await params.dispatcherOptions.onReplyStart?.();
         await params.dispatcherOptions.deliver({ text: "replying now" }, { kind: "final" });
-        await params.dispatcherOptions.onIdle?.();
+        params.dispatcherOptions.onIdle?.();
         return EMPTY_DISPATCH_RESULT;
       });
 

extensions/bluebubbles/src/setup-surface.ts

@@ -83,7 +83,9 @@ function validateBlueBubblesServerUrlInput(value: unknown): string | undefined {
   }
   try {
     const normalized = normalizeBlueBubblesServerUrl(trimmed);
-    new URL(normalized);
+    if (!URL.canParse(normalized)) {
+      return "Invalid URL format";
+    }
     return undefined;
   } catch {
     return "Invalid URL format";
@@ -109,7 +111,7 @@ function applyBlueBubblesSetupPatch(
 }
 
 function validateBlueBubblesWebhookPath(value: string): string | undefined {
-  const trimmed = String(value ?? "").trim();
+  const trimmed = value.trim();
   if (!trimmed) {
     return "Required";
   }
@@ -222,7 +224,7 @@ export const blueBubblesSetupWizard: ChannelSetupWizard = {
       currentValue: ({ cfg, accountId }) =>
         normalizeOptionalString(resolveBlueBubblesAccount({ cfg, accountId }).config.serverUrl),
       validate: ({ value }) => validateBlueBubblesServerUrlInput(value),
-      normalizeValue: ({ value }) => String(value).trim(),
+      normalizeValue: ({ value }) => value.trim(),
       applySet: async ({ cfg, accountId, value }) =>
         applyBlueBubblesSetupPatch(cfg, accountId, {
           serverUrl: value,
@@ -241,7 +243,7 @@ export const blueBubblesSetupWizard: ChannelSetupWizard = {
       shouldPrompt: ({ credentialValues }) =>
         credentialValues[CONFIGURE_CUSTOM_WEBHOOK_FLAG] === "1",
       validate: ({ value }) => validateBlueBubblesWebhookPath(value),
-      normalizeValue: ({ value }) => String(value).trim(),
+      normalizeValue: ({ value }) => value.trim(),
       applySet: async ({ cfg, accountId, value }) =>
         applyBlueBubblesSetupPatch(cfg, accountId, {
           webhookPath: value,

extensions/browser/index.test.ts

@@ -65,7 +65,7 @@ describe("browser plugin", () => {
 
   it("forwards per-session browser options into the tool factory", async () => {
     const { api, registerTool } = createApi();
-    await registerBrowserPlugin(api);
+    registerBrowserPlugin(api);
 
     const tool = registerTool.mock.calls[0]?.[0];
     if (typeof tool !== "function") {

extensions/browser/src/browser-tool.ts

@@ -286,7 +286,7 @@ async function callBrowserProxy(params: {
       ? Math.max(1, Math.floor(params.timeoutMs))
       : DEFAULT_BROWSER_PROXY_TIMEOUT_MS;
   const gatewayTimeoutMs = proxyTimeoutMs + BROWSER_PROXY_GATEWAY_TIMEOUT_SLACK_MS;
-  const payload = await browserToolDeps.callGatewayTool<{ payloadJSON?: string; payload?: string }>(
+  const payload = await browserToolDeps.callGatewayTool(
     "node.invoke",
     { timeoutMs: gatewayTimeoutMs },
     {

extensions/browser/src/browser/bridge-server.auth.test.ts

@@ -83,10 +83,12 @@ describe("startBrowserBridgeServer auth", () => {
   });
 
   it("serves noVNC bootstrap html without leaking password in Location header", async () => {
+    let resolveCalls = 0;
     const bridge = await startBrowserBridgeServer({
       resolved: buildResolvedConfig(),
       authToken: "secret-token",
       resolveSandboxNoVncToken: (token) => {
+        resolveCalls += 1;
         if (token !== "valid-token") {
           return null;
         }
@@ -95,8 +97,15 @@ describe("startBrowserBridgeServer auth", () => {
     });
     servers.push({ stop: () => stopBrowserBridgeServer(bridge.server) });
 
-    const res = await fetch(`${bridge.baseUrl}/sandbox/novnc?token=valid-token`);
+    const unauth = await fetch(`${bridge.baseUrl}/sandbox/novnc?token=valid-token`);
+    expect(unauth.status).toBe(401);
+    expect(resolveCalls).toBe(0);
+
+    const res = await fetch(`${bridge.baseUrl}/sandbox/novnc?token=valid-token`, {
+      headers: { Authorization: "Bearer secret-token" },
+    });
     expect(res.status).toBe(200);
+    expect(resolveCalls).toBe(1);
     expect(res.headers.get("location")).toBeNull();
     expect(res.headers.get("cache-control")).toContain("no-store");
     expect(res.headers.get("referrer-policy")).toBe("no-referrer");

extensions/browser/src/browser/bridge-server.ts

@@ -13,6 +13,7 @@ import {
   type ProfileContext,
 } from "./server-context.js";
 import {
+  hasVerifiedBrowserAuth,
   installBrowserAuthMiddleware,
   installBrowserCommonMiddleware,
 } from "./server-middleware.js";
@@ -76,8 +77,19 @@ export async function startBrowserBridgeServer(params: {
   const app = express();
   installBrowserCommonMiddleware(app);
 
+  const authToken = normalizeOptionalString(params.authToken);
+  const authPassword = normalizeOptionalString(params.authPassword);
+  if (!authToken && !authPassword) {
+    throw new Error("bridge server requires auth (authToken/authPassword missing)");
+  }
+  installBrowserAuthMiddleware(app, { token: authToken, password: authPassword });
+
   if (params.resolveSandboxNoVncToken) {
     app.get("/sandbox/novnc", (req, res) => {
+      if (!hasVerifiedBrowserAuth(req)) {
+        res.status(401).send("Unauthorized");
+        return;
+      }
       res.setHeader("Cache-Control", "no-store, no-cache, must-revalidate, proxy-revalidate");
       res.setHeader("Pragma", "no-cache");
       res.setHeader("Expires", "0");
@@ -96,13 +108,6 @@ export async function startBrowserBridgeServer(params: {
     });
   }
 
-  const authToken = normalizeOptionalString(params.authToken);
-  const authPassword = normalizeOptionalString(params.authPassword);
-  if (!authToken && !authPassword) {
-    throw new Error("bridge server requires auth (authToken/authPassword missing)");
-  }
-  installBrowserAuthMiddleware(app, { token: authToken, password: authPassword });
-
   const state: BrowserServerState = {
     server: null as unknown as Server,
     port,

extensions/browser/src/browser/cdp.helpers.test.ts

@@ -0,0 +1,65 @@
+import { afterEach, describe, expect, it, vi } from "vitest";
+
+const fetchWithSsrFGuardMock = vi.hoisted(() => vi.fn());
+
+vi.mock("openclaw/plugin-sdk/ssrf-runtime", async (importOriginal) => {
+  const actual = await importOriginal<typeof import("openclaw/plugin-sdk/ssrf-runtime")>();
+  return {
+    ...actual,
+    fetchWithSsrFGuard: (...args: unknown[]) => fetchWithSsrFGuardMock(...args),
+  };
+});
+
+import { fetchJson, fetchOk } from "./cdp.helpers.js";
+
+describe("cdp helpers", () => {
+  afterEach(() => {
+    fetchWithSsrFGuardMock.mockReset();
+  });
+
+  it("releases guarded CDP fetches after the response body is consumed", async () => {
+    const release = vi.fn(async () => {});
+    const json = vi.fn(async () => {
+      expect(release).not.toHaveBeenCalled();
+      return { ok: true };
+    });
+    fetchWithSsrFGuardMock.mockResolvedValueOnce({
+      response: {
+        ok: true,
+        status: 200,
+        json,
+      },
+      release,
+    });
+
+    await expect(
+      fetchJson("http://127.0.0.1:9222/json/version", 250, undefined, {
+        dangerouslyAllowPrivateNetwork: false,
+        allowedHostnames: ["127.0.0.1"],
+      }),
+    ).resolves.toEqual({ ok: true });
+
+    expect(json).toHaveBeenCalledTimes(1);
+    expect(release).toHaveBeenCalledTimes(1);
+  });
+
+  it("releases guarded CDP fetches for bodyless requests", async () => {
+    const release = vi.fn(async () => {});
+    fetchWithSsrFGuardMock.mockResolvedValueOnce({
+      response: {
+        ok: true,
+        status: 200,
+      },
+      release,
+    });
+
+    await expect(
+      fetchOk("http://127.0.0.1:9222/json/close/TARGET_1", 250, undefined, {
+        dangerouslyAllowPrivateNetwork: false,
+        allowedHostnames: ["127.0.0.1"],
+      }),
+    ).resolves.toBeUndefined();
+
+    expect(release).toHaveBeenCalledTimes(1);
+  });
+});

extensions/browser/src/browser/cdp.helpers.ts

@@ -1,11 +1,17 @@
+import { fetchWithSsrFGuard } from "openclaw/plugin-sdk/ssrf-runtime";
 import { normalizeLowercaseStringOrEmpty } from "openclaw/plugin-sdk/text-runtime";
 import WebSocket from "ws";
 import { isLoopbackHost } from "../gateway/net.js";
-import { type SsrFPolicy, resolvePinnedHostnameWithPolicy } from "../infra/net/ssrf.js";
+import {
+  SsrFBlockedError,
+  type SsrFPolicy,
+  resolvePinnedHostnameWithPolicy,
+} from "../infra/net/ssrf.js";
 import { rawDataToString } from "../infra/ws.js";
 import { redactSensitiveText } from "../logging/redact.js";
 import { getDirectAgentForCdp, withNoProxyForCdpUrl } from "./cdp-proxy-bypass.js";
 import { CDP_HTTP_REQUEST_TIMEOUT_MS, CDP_WS_HANDSHAKE_TIMEOUT_MS } from "./cdp-timeouts.js";
+import { BrowserCdpEndpointBlockedError } from "./errors.js";
 import { resolveBrowserRateLimitMessage } from "./rate-limit-message.js";
 
 export { isLoopbackHost };
@@ -62,9 +68,13 @@ export async function assertCdpEndpointAllowed(
   if (!["http:", "https:", "ws:", "wss:"].includes(parsed.protocol)) {
     throw new Error(`Invalid CDP URL protocol: ${parsed.protocol.replace(":", "")}`);
   }
-  await resolvePinnedHostnameWithPolicy(parsed.hostname, {
-    policy: ssrfPolicy,
-  });
+  try {
+    await resolvePinnedHostnameWithPolicy(parsed.hostname, {
+      policy: ssrfPolicy,
+    });
+  } catch (error) {
+    throw new BrowserCdpEndpointBlockedError({ cause: error });
+  }
 }
 
 export function redactCdpUrl(cdpUrl: string | null | undefined): string | null | undefined {
@@ -152,6 +162,11 @@ export function normalizeCdpHttpBaseForJsonEndpoints(cdpUrl: string): string {
   }
 }
 
+type CdpFetchResult = {
+  response: Response;
+  release: () => Promise<void>;
+};
+
 function createCdpSender(ws: WebSocket) {
   let nextId = 1;
   const pending = new Map<number, Pending>();
@@ -217,23 +232,47 @@ export async function fetchJson<T>(
   url: string,
   timeoutMs = CDP_HTTP_REQUEST_TIMEOUT_MS,
   init?: RequestInit,
+  ssrfPolicy?: SsrFPolicy,
 ): Promise<T> {
-  const res = await fetchCdpChecked(url, timeoutMs, init);
-  return (await res.json()) as T;
+  const { response, release } = await fetchCdpChecked(url, timeoutMs, init, ssrfPolicy);
+  try {
+    return (await response.json()) as T;
+  } finally {
+    await release();
+  }
 }
 
 export async function fetchCdpChecked(
   url: string,
   timeoutMs = CDP_HTTP_REQUEST_TIMEOUT_MS,
   init?: RequestInit,
-): Promise<Response> {
+  ssrfPolicy?: SsrFPolicy,
+): Promise<CdpFetchResult> {
   const ctrl = new AbortController();
   const t = setTimeout(ctrl.abort.bind(ctrl), timeoutMs);
+  let guardedRelease: (() => Promise<void>) | undefined;
+  let released = false;
+  const release = async () => {
+    if (released) {
+      return;
+    }
+    released = true;
+    clearTimeout(t);
+    await guardedRelease?.();
+  };
   try {
     const headers = getHeadersWithAuth(url, (init?.headers as Record<string, string>) || {});
-    const res = await withNoProxyForCdpUrl(url, () =>
-      fetch(url, { ...init, headers, signal: ctrl.signal }),
-    );
+    const res = await withNoProxyForCdpUrl(url, async () => {
+      const guarded = await fetchWithSsrFGuard({
+        url,
+        init: { ...init, headers },
+        signal: ctrl.signal,
+        policy: ssrfPolicy ?? { allowPrivateNetwork: true },
+        auditContext: "browser-cdp",
+      });
+      guardedRelease = guarded.release;
+      return guarded.response;
+    });
     if (!res.ok) {
       if (res.status === 429) {
         // Do not reflect upstream response text into the error surface (log/agent injection risk)
@@ -241,9 +280,13 @@ export async function fetchCdpChecked(
       }
       throw new Error(`HTTP ${res.status}`);
     }
-    return res;
-  } finally {
-    clearTimeout(t);
+    return { response: res, release };
+  } catch (error) {
+    await release();
+    if (error instanceof SsrFBlockedError) {
+      throw new BrowserCdpEndpointBlockedError({ cause: error });
+    }
+    throw error;
   }
 }
 
@@ -251,8 +294,10 @@ export async function fetchOk(
   url: string,
   timeoutMs = CDP_HTTP_REQUEST_TIMEOUT_MS,
   init?: RequestInit,
+  ssrfPolicy?: SsrFPolicy,
 ): Promise<void> {
-  await fetchCdpChecked(url, timeoutMs, init);
+  const { release } = await fetchCdpChecked(url, timeoutMs, init, ssrfPolicy);
+  await release();
 }
 
 export function openCdpWebSocket(

extensions/browser/src/browser/cdp.test.ts

@@ -1,13 +1,29 @@
 import { createServer } from "node:http";
-import { afterEach, describe, expect, it, vi } from "vitest";
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
 import { type WebSocket, WebSocketServer } from "ws";
 import { SsrFBlockedError } from "../infra/net/ssrf.js";
 import { rawDataToString } from "../infra/ws.js";
 import { isWebSocketUrl } from "./cdp.helpers.js";
 import { createTargetViaCdp, evaluateJavaScript, normalizeCdpWsUrl, snapshotAria } from "./cdp.js";
 import { parseHttpUrl } from "./config.js";
+import { BrowserCdpEndpointBlockedError } from "./errors.js";
 import { InvalidBrowserNavigationUrlError } from "./navigation-guard.js";
 
+vi.mock("openclaw/plugin-sdk/browser-security-runtime", async () => {
+  const actual = await vi.importActual<
+    typeof import("openclaw/plugin-sdk/browser-security-runtime")
+  >("openclaw/plugin-sdk/browser-security-runtime");
+  const lookupFn = async (_hostname: string, options?: { all?: boolean }) => {
+    const result = { address: "93.184.216.34", family: 4 };
+    return options?.all === true ? [result] : result;
+  };
+  return {
+    ...actual,
+    resolvePinnedHostnameWithPolicy: (hostname: string, params: object = {}) =>
+      actual.resolvePinnedHostnameWithPolicy(hostname, { ...params, lookupFn: lookupFn as never }),
+  };
+});
+
 describe("cdp", () => {
   let httpServer: ReturnType<typeof createServer> | null = null;
   let wsServer: WebSocketServer | null = null;
@@ -56,6 +72,7 @@ describe("cdp", () => {
   };
 
   afterEach(async () => {
+    vi.unstubAllEnvs();
     await new Promise<void>((resolve) => {
       if (!httpServer) {
         return resolve();
@@ -185,6 +202,22 @@ describe("cdp", () => {
     }
   });
 
+  it("blocks hostname navigation targets when strict SSRF policy is configured", async () => {
+    const fetchSpy = vi.spyOn(globalThis, "fetch");
+    try {
+      await expect(
+        createTargetViaCdp({
+          cdpUrl: "http://127.0.0.1:9222",
+          url: "https://example.com",
+          ssrfPolicy: { dangerouslyAllowPrivateNetwork: false },
+        }),
+      ).rejects.toBeInstanceOf(InvalidBrowserNavigationUrlError);
+      expect(fetchSpy).not.toHaveBeenCalled();
+    } finally {
+      fetchSpy.mockRestore();
+    }
+  });
+
   it("blocks unsupported non-network navigation URLs", async () => {
     const fetchSpy = vi.spyOn(globalThis, "fetch");
     try {
@@ -235,39 +268,39 @@ describe("cdp", () => {
     await expect(
       createTargetViaCdp({
         cdpUrl: `http://127.0.0.1:${httpPort}`,
-        url: "https://example.com",
+        url: "https://93.184.216.34",
         ssrfPolicy: {
           dangerouslyAllowPrivateNetwork: false,
           allowedHostnames: ["127.0.0.1"],
         },
       }),
-    ).rejects.toBeInstanceOf(SsrFBlockedError);
+    ).rejects.toBeInstanceOf(BrowserCdpEndpointBlockedError);
   });
 
   it("blocks the initial /json/version fetch when the cdpUrl host is outside strict SSRF policy", async () => {
     await expect(
       createTargetViaCdp({
         cdpUrl: "http://169.254.169.254:9222",
-        url: "https://example.com",
+        url: "https://93.184.216.34",
         ssrfPolicy: {
           dangerouslyAllowPrivateNetwork: false,
           allowedHostnames: ["127.0.0.1"],
         },
       }),
-    ).rejects.toBeInstanceOf(SsrFBlockedError);
+    ).rejects.toBeInstanceOf(BrowserCdpEndpointBlockedError);
   });
 
   it("blocks direct websocket cdp urls outside strict SSRF policy", async () => {
     await expect(
       createTargetViaCdp({
         cdpUrl: "ws://169.254.169.254:9222/devtools/browser/PIVOT",
-        url: "https://example.com",
+        url: "https://93.184.216.34",
         ssrfPolicy: {
           dangerouslyAllowPrivateNetwork: false,
           allowedHostnames: ["127.0.0.1"],
         },
       }),
-    ).rejects.toBeInstanceOf(SsrFBlockedError);
+    ).rejects.toBeInstanceOf(BrowserCdpEndpointBlockedError);
   });
 
   it("evaluates javascript via CDP", async () => {
@@ -470,3 +503,17 @@ describe("parseHttpUrl with WebSocket protocols", () => {
     expect(() => parseHttpUrl("file:///etc/passwd", "test")).toThrow("must be http(s) or ws(s)");
   });
 });
+const proxyEnvKeys = [
+  "ALL_PROXY",
+  "all_proxy",
+  "HTTP_PROXY",
+  "http_proxy",
+  "HTTPS_PROXY",
+  "https_proxy",
+] as const;
+
+beforeEach(() => {
+  for (const key of proxyEnvKeys) {
+    vi.stubEnv(key, "");
+  }
+});

extensions/browser/src/browser/cdp.ts

@@ -78,8 +78,8 @@ export async function captureScreenshot(opts: {
         contentSize?: { width?: number; height?: number };
       };
       const size = metrics?.cssContentSize ?? metrics?.contentSize;
-      const contentWidth = Number(size?.width ?? 0);
-      const contentHeight = Number(size?.height ?? 0);
+      const contentWidth = size?.width ?? 0;
+      const contentHeight = size?.height ?? 0;
       if (contentWidth > 0 && contentHeight > 0) {
         const vpResult = (await send("Runtime.evaluate", {
           expression:
@@ -91,14 +91,14 @@ export async function captureScreenshot(opts: {
           };
         };
         const v = vpResult?.result?.value;
-        const currentW = Number(v?.w ?? 0);
-        const currentH = Number(v?.h ?? 0);
+        const currentW = v?.w ?? 0;
+        const currentH = v?.h ?? 0;
         savedVp = {
           w: currentW,
           h: currentH,
-          dpr: Number(v?.dpr ?? 1),
-          sw: Number(v?.sw ?? currentW),
-          sh: Number(v?.sh ?? currentH),
+          dpr: v?.dpr ?? 1,
+          sw: v?.sw ?? currentW,
+          sh: v?.sh ?? currentH,
         };
         // mobile: false is the safe default — CDP provides no way to query
         // the active mobile flag, and inferring from navigator.maxTouchPoints
@@ -148,11 +148,7 @@ export async function captureScreenshot(opts: {
             returnByValue: true,
           })) as { result?: { value?: { w?: number; h?: number; dpr?: number } } };
           const p = postResult?.result?.value;
-          if (
-            Number(p?.w) !== savedVp.w ||
-            Number(p?.h) !== savedVp.h ||
-            Number(p?.dpr) !== savedVp.dpr
-          ) {
+          if (p?.w !== savedVp.w || p?.h !== savedVp.h || p?.dpr !== savedVp.dpr) {
             await send("Emulation.setDeviceMetricsOverride", {
               width: savedVp.w,
               height: savedVp.h,
@@ -187,12 +183,13 @@ export async function createTargetViaCdp(opts: {
     wsUrl = opts.cdpUrl;
   } else {
     // Standard HTTP(S) CDP endpoint — discover WebSocket URL via /json/version.
-    await assertCdpEndpointAllowed(opts.cdpUrl, opts.ssrfPolicy);
     const version = await fetchJson<{ webSocketDebuggerUrl?: string }>(
       appendCdpPath(opts.cdpUrl, "/json/version"),
       1500,
+      undefined,
+      opts.ssrfPolicy,
     );
-    const wsUrlRaw = String(version?.webSocketDebuggerUrl ?? "").trim();
+    const wsUrlRaw = version?.webSocketDebuggerUrl?.trim() ?? "";
     wsUrl = wsUrlRaw ? normalizeCdpWsUrl(wsUrlRaw, opts.cdpUrl) : "";
     if (!wsUrl) {
       throw new Error("CDP /json/version missing webSocketDebuggerUrl");
@@ -204,7 +201,7 @@ export async function createTargetViaCdp(opts: {
     const created = (await send("Target.createTarget", { url: opts.url })) as {
       targetId?: string;
     };
-    const targetId = String(created?.targetId ?? "").trim();
+    const targetId = created?.targetId?.trim() ?? "";
     if (!targetId) {
       throw new Error("CDP Target.createTarget returned no targetId");
     }

extensions/browser/src/browser/chrome.executables.ts

@@ -691,7 +691,7 @@ export function readBrowserVersion(executablePath: string): string | null {
 }
 
 export function parseBrowserMajorVersion(rawVersion: string | null | undefined): number | null {
-  const matches = [...String(rawVersion ?? "").matchAll(CHROME_VERSION_RE)];
+  const matches = [...(rawVersion ?? "").matchAll(CHROME_VERSION_RE)];
   const match = matches.at(-1);
   if (!match?.[1]) {
     return null;

extensions/browser/src/browser/chrome.test.ts

@@ -6,7 +6,6 @@ import os from "node:os";
 import path from "node:path";
 import { afterAll, afterEach, beforeAll, beforeEach, describe, expect, it, vi } from "vitest";
 import { WebSocketServer } from "ws";
-import { SsrFBlockedError } from "../infra/net/ssrf.js";
 import {
   decorateOpenClawProfile,
   ensureProfileCleanExit,
@@ -22,6 +21,7 @@ import {
   DEFAULT_OPENCLAW_BROWSER_COLOR,
   DEFAULT_OPENCLAW_BROWSER_PROFILE_NAME,
 } from "./constants.js";
+import { BrowserCdpEndpointBlockedError } from "./errors.js";
 
 type StopChromeTarget = Parameters<typeof stopOpenClawChrome>[0];
 
@@ -357,7 +357,7 @@ describe("browser chrome helpers", () => {
           dangerouslyAllowPrivateNetwork: false,
           allowedHostnames: ["127.0.0.1"],
         }),
-      ).rejects.toBeInstanceOf(SsrFBlockedError);
+      ).rejects.toBeInstanceOf(BrowserCdpEndpointBlockedError);
     } finally {
       await new Promise<void>((resolve) => server.close(() => resolve()));
     }

extensions/browser/src/browser/chrome.ts

@@ -171,14 +171,22 @@ async function fetchChromeVersion(
   const ctrl = new AbortController();
   const t = setTimeout(ctrl.abort.bind(ctrl), timeoutMs);
   try {
-    await assertCdpEndpointAllowed(cdpUrl, ssrfPolicy);
     const versionUrl = appendCdpPath(cdpUrl, "/json/version");
-    const res = await fetchCdpChecked(versionUrl, timeoutMs, { signal: ctrl.signal });
-    const data = (await res.json()) as ChromeVersion;
-    if (!data || typeof data !== "object") {
-      return null;
+    const { response, release } = await fetchCdpChecked(
+      versionUrl,
+      timeoutMs,
+      { signal: ctrl.signal },
+      ssrfPolicy,
+    );
+    try {
+      const data = (await response.json()) as ChromeVersion;
+      if (!data || typeof data !== "object") {
+        return null;
+      }
+      return data;
+    } finally {
+      await release();
     }
-    return data;
   } catch {
     return null;
   } finally {

extensions/browser/src/browser/client-fetch.loopback-auth.test.ts

@@ -1,6 +1,42 @@
 import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
 import type { BrowserDispatchResponse } from "./routes/dispatcher.js";
 
+vi.mock("openclaw/plugin-sdk/browser-security-runtime", async () => {
+  const actual = await vi.importActual<
+    typeof import("openclaw/plugin-sdk/browser-security-runtime")
+  >("openclaw/plugin-sdk/browser-security-runtime");
+  const lookupFn = async (_hostname: string, options?: { all?: boolean }) => {
+    const result = { address: "93.184.216.34", family: 4 };
+    return options?.all === true ? [result] : result;
+  };
+  return {
+    ...actual,
+    resolvePinnedHostnameWithPolicy: (hostname: string, params: object = {}) =>
+      actual.resolvePinnedHostnameWithPolicy(hostname, { ...params, lookupFn: lookupFn as never }),
+  };
+});
+
+vi.mock("openclaw/plugin-sdk/ssrf-runtime", async () => {
+  const actual = await vi.importActual<typeof import("openclaw/plugin-sdk/ssrf-runtime")>(
+    "openclaw/plugin-sdk/ssrf-runtime",
+  );
+  return {
+    ...actual,
+    fetchWithSsrFGuard: async (params: {
+      url: string;
+      init?: RequestInit;
+      signal?: AbortSignal;
+    }) => ({
+      response: await fetch(params.url, {
+        ...params.init,
+        signal: params.signal,
+      }),
+      finalUrl: params.url,
+      release: async () => {},
+    }),
+  };
+});
+
 function okDispatchResponse(): BrowserDispatchResponse {
   return { status: 200, body: { ok: true } };
 }
@@ -87,6 +123,16 @@ async function expectThrownBrowserFetchError(
 describe("fetchBrowserJson loopback auth", () => {
   beforeEach(() => {
     vi.restoreAllMocks();
+    for (const key of [
+      "ALL_PROXY",
+      "all_proxy",
+      "HTTP_PROXY",
+      "http_proxy",
+      "HTTPS_PROXY",
+      "https_proxy",
+    ]) {
+      vi.stubEnv(key, "");
+    }
     vi.stubEnv("OPENCLAW_GATEWAY_TOKEN", "loopback-token");
     mocks.loadConfig.mockClear();
     mocks.loadConfig.mockReturnValue({

extensions/browser/src/browser/config.test.ts

@@ -307,11 +307,9 @@ describe("browser config", () => {
     });
   });
 
-  it("defaults browser SSRF policy to trusted-network mode", () => {
+  it("defaults browser SSRF policy to strict mode when unset", () => {
     const resolved = resolveBrowserConfig({});
-    expect(resolved.ssrfPolicy).toEqual({
-      dangerouslyAllowPrivateNetwork: true,
-    });
+    expect(resolved.ssrfPolicy).toEqual({});
   });
 
   it("supports explicit strict mode by disabling private network access", () => {
@@ -323,6 +321,19 @@ describe("browser config", () => {
     expect(resolved.ssrfPolicy).toEqual({});
   });
 
+  it("keeps allowlist-only browser SSRF policy strict by default", () => {
+    const resolved = resolveBrowserConfig({
+      ssrfPolicy: {
+        allowedHostnames: ["example.com"],
+        hostnameAllowlist: ["*.example.com"],
+      },
+    } as unknown as BrowserConfig);
+    expect(resolved.ssrfPolicy).toEqual({
+      allowedHostnames: ["example.com"],
+      hostnameAllowlist: ["*.example.com"],
+    });
+  });
+
   it("resolves existing-session profiles without cdpPort or cdpUrl", () => {
     const resolved = resolveBrowserConfig({
       profiles: {

extensions/browser/src/browser/config.ts

@@ -42,6 +42,14 @@ export {
 export type { BrowserControlAuth };
 export { parseBrowserHttpUrl as parseHttpUrl };
 
+type BrowserSsrFPolicyCompat = NonNullable<BrowserConfig["ssrfPolicy"]> & {
+  /**
+   * Legacy raw-config alias. Keep it out of the public BrowserConfig type while
+   * still accepting old user files until doctor rewrites them.
+   */
+  allowPrivateNetwork?: boolean;
+};
+
 export type ResolvedBrowserConfig = {
   enabled: boolean;
   evaluateEnabled: boolean;
@@ -119,9 +127,7 @@ function resolveCdpPortRangeStart(
 const normalizeStringList = normalizeOptionalTrimmedStringList;
 
 function resolveBrowserSsrFPolicy(cfg: BrowserConfig | undefined): SsrFPolicy | undefined {
-  const rawPolicy = cfg?.ssrfPolicy as
-    | (BrowserConfig["ssrfPolicy"] & { allowPrivateNetwork?: boolean })
-    | undefined;
+  const rawPolicy = cfg?.ssrfPolicy as BrowserSsrFPolicyCompat | undefined;
   const allowPrivateNetwork = rawPolicy?.allowPrivateNetwork;
   const dangerouslyAllowPrivateNetwork = rawPolicy?.dangerouslyAllowPrivateNetwork;
   const allowedHostnames = normalizeStringList(rawPolicy?.allowedHostnames);
@@ -129,9 +135,7 @@ function resolveBrowserSsrFPolicy(cfg: BrowserConfig | undefined): SsrFPolicy |
   const hasExplicitPrivateSetting =
     allowPrivateNetwork !== undefined || dangerouslyAllowPrivateNetwork !== undefined;
   const resolvedAllowPrivateNetwork =
-    dangerouslyAllowPrivateNetwork === true ||
-    allowPrivateNetwork === true ||
-    !hasExplicitPrivateSetting;
+    dangerouslyAllowPrivateNetwork === true || allowPrivateNetwork === true;
 
   if (
     !resolvedAllowPrivateNetwork &&
@@ -139,7 +143,9 @@ function resolveBrowserSsrFPolicy(cfg: BrowserConfig | undefined): SsrFPolicy |
     !allowedHostnames &&
     !hostnameAllowlist
   ) {
-    return undefined;
+    // Keep the default policy object present so CDP guards still enforce
+    // fail-closed private-network checks on unconfigured installs.
+    return {};
   }
 
   return {

extensions/browser/src/browser/control-auth.auto-token.test.ts

@@ -9,7 +9,7 @@ const mocks = vi.hoisted(() => ({
     ({
       authConfig,
     }: {
-      authConfig?: NonNullable<NonNullable<OpenClawConfig["gateway"]>["auth"]> | undefined;
+      authConfig?: NonNullable<NonNullable<OpenClawConfig["gateway"]>["auth"]>;
     }) => {
       const token =
         typeof authConfig?.token === "string"
@@ -58,6 +58,14 @@ vi.mock("../gateway/auth.js", () => ({
   resolveGatewayAuth: mocks.resolveGatewayAuth,
 }));
 
+function readPersistedConfig(): OpenClawConfig {
+  const persistedCfg = mocks.writeConfigFile.mock.calls[0]?.[0];
+  if (!persistedCfg) {
+    throw new Error("expected persisted config");
+  }
+  return persistedCfg;
+}
+
 let ensureBrowserControlAuth: typeof import("./control-auth.js").ensureBrowserControlAuth;
 
 describe("ensureBrowserControlAuth", () => {
@@ -176,7 +184,7 @@ describe("ensureBrowserControlAuth", () => {
     expect(result.auth.token).toBe(result.generatedToken);
     expect(result.auth.password).toBeUndefined();
     expect(mocks.writeConfigFile).toHaveBeenCalledTimes(1);
-    const persistedCfg = mocks.writeConfigFile.mock.calls[0]?.[0] as OpenClawConfig | undefined;
+    const persistedCfg = readPersistedConfig();
     expect(persistedCfg?.gateway?.auth?.mode).toBe("none");
     expect(persistedCfg?.gateway?.auth?.token).toBe(result.generatedToken);
     expect(mocks.ensureGatewayStartupAuth).not.toHaveBeenCalled();
@@ -223,7 +231,7 @@ describe("ensureBrowserControlAuth", () => {
     expect(result.auth.token).toBe(result.generatedToken);
     expect(result.auth.password).toBeUndefined();
     expect(mocks.writeConfigFile).toHaveBeenCalledTimes(1);
-    const persistedCfg = mocks.writeConfigFile.mock.calls[0]?.[0] as OpenClawConfig | undefined;
+    const persistedCfg = readPersistedConfig();
     expect(persistedCfg?.gateway?.auth?.mode).toBe("none");
     expect(persistedCfg?.gateway?.auth?.token).toBe(result.generatedToken);
     expect(mocks.ensureGatewayStartupAuth).not.toHaveBeenCalled();
@@ -246,7 +254,7 @@ describe("ensureBrowserControlAuth", () => {
     expect(result.auth.password).toBe(result.generatedToken);
     expect(result.auth.token).toBeUndefined();
     expect(mocks.writeConfigFile).toHaveBeenCalledTimes(1);
-    const persistedCfg = mocks.writeConfigFile.mock.calls[0]?.[0] as OpenClawConfig | undefined;
+    const persistedCfg = readPersistedConfig();
     expect(persistedCfg?.gateway?.auth?.mode).toBe("trusted-proxy");
     expect(persistedCfg?.gateway?.auth?.password).toBe(result.generatedToken);
     expect(mocks.ensureGatewayStartupAuth).not.toHaveBeenCalled();
@@ -273,7 +281,7 @@ describe("ensureBrowserControlAuth", () => {
     expect(result.auth.password).toBe(result.generatedToken);
     expect(result.auth.token).toBeUndefined();
     expect(mocks.writeConfigFile).toHaveBeenCalledTimes(1);
-    const persistedCfg = mocks.writeConfigFile.mock.calls[0]?.[0] as OpenClawConfig | undefined;
+    const persistedCfg = readPersistedConfig();
     expect(persistedCfg?.gateway?.auth?.mode).toBe("trusted-proxy");
     expect(persistedCfg?.gateway?.auth?.password).toBe(result.generatedToken);
     expect(mocks.ensureGatewayStartupAuth).not.toHaveBeenCalled();

extensions/browser/src/browser/errors.test.ts

@@ -1,5 +1,12 @@
 import { describe, expect, it } from "vitest";
-import { BrowserValidationError, toBrowserErrorResponse } from "./errors.js";
+import { SsrFBlockedError } from "../infra/net/ssrf.js";
+import {
+  BROWSER_ENDPOINT_BLOCKED_MESSAGE,
+  BROWSER_NAVIGATION_BLOCKED_MESSAGE,
+  BrowserCdpEndpointBlockedError,
+  BrowserValidationError,
+  toBrowserErrorResponse,
+} from "./errors.js";
 
 describe("browser error mapping", () => {
   it("maps blocked browser targets to conflict responses", () => {
@@ -20,4 +27,22 @@ describe("browser error mapping", () => {
       message: "bad input",
     });
   });
+
+  it("sanitizes navigation-target SSRF policy errors without leaking raw policy details", () => {
+    expect(
+      toBrowserErrorResponse(
+        new SsrFBlockedError("Blocked hostname or private/internal/special-use IP address"),
+      ),
+    ).toEqual({
+      status: 400,
+      message: BROWSER_NAVIGATION_BLOCKED_MESSAGE,
+    });
+  });
+
+  it("maps CDP endpoint policy blocks to a distinct endpoint-scoped message", () => {
+    expect(toBrowserErrorResponse(new BrowserCdpEndpointBlockedError())).toEqual({
+      status: 400,
+      message: BROWSER_ENDPOINT_BLOCKED_MESSAGE,
+    });
+  });
 });

extensions/browser/src/browser/errors.ts

@@ -1,6 +1,9 @@
 import { SsrFBlockedError } from "../infra/net/ssrf.js";
 import { InvalidBrowserNavigationUrlError } from "./navigation-guard.js";
 
+export const BROWSER_ENDPOINT_BLOCKED_MESSAGE = "browser endpoint blocked by policy";
+export const BROWSER_NAVIGATION_BLOCKED_MESSAGE = "browser navigation blocked by policy";
+
 export class BrowserError extends Error {
   status: number;
 
@@ -11,6 +14,18 @@ export class BrowserError extends Error {
   }
 }
 
+/**
+ * Raised when a browser CDP endpoint (the cdpUrl itself) fails the
+ * configured SSRF policy. Distinct from a blocked navigation target so
+ * callers see "fix your browser endpoint config" rather than "fix your
+ * navigation URL".
+ */
+export class BrowserCdpEndpointBlockedError extends BrowserError {
+  constructor(options?: ErrorOptions) {
+    super(BROWSER_ENDPOINT_BLOCKED_MESSAGE, 400, options);
+  }
+}
+
 export class BrowserValidationError extends BrowserError {
   constructor(message: string, options?: ErrorOptions) {
     super(message, 400, options);
@@ -76,7 +91,12 @@ export function toBrowserErrorResponse(err: unknown): {
     return { status: 409, message: err.message };
   }
   if (err instanceof SsrFBlockedError) {
-    return { status: 400, message: err.message };
+    // SsrFBlockedError from this point is from a navigation-target check
+    // (assertBrowserNavigationAllowed / resolvePinnedHostnameWithPolicy on a
+    // requested URL). CDP endpoint blocks are rethrown as
+    // BrowserCdpEndpointBlockedError by assertCdpEndpointAllowed and handled
+    // by the BrowserError branch above.
+    return { status: 400, message: BROWSER_NAVIGATION_BLOCKED_MESSAGE };
   }
   if (
     err instanceof InvalidBrowserNavigationUrlError ||

extensions/browser/src/browser/navigation-guard.test.ts

@@ -116,6 +116,85 @@ describe("browser navigation guard", () => {
     expect(lookupFn).toHaveBeenCalledWith("example.com", { all: true });
   });
 
+  it("blocks hostname navigation when strict SSRF policy is explicitly configured", async () => {
+    const lookupFn = createLookupFn("93.184.216.34");
+    await expect(
+      assertBrowserNavigationAllowed({
+        url: "https://example.com",
+        lookupFn,
+        ssrfPolicy: { dangerouslyAllowPrivateNetwork: false },
+      }),
+    ).rejects.toThrow(/dns rebinding protections are unavailable/i);
+    expect(lookupFn).not.toHaveBeenCalled();
+  });
+
+  it("allows explicitly allowed hostnames in strict mode", async () => {
+    const lookupFn = createLookupFn("93.184.216.34");
+    await expect(
+      assertBrowserNavigationAllowed({
+        url: "https://agent.internal",
+        lookupFn,
+        ssrfPolicy: {
+          dangerouslyAllowPrivateNetwork: false,
+          allowedHostnames: ["agent.internal"],
+        },
+      }),
+    ).resolves.toBeUndefined();
+  });
+
+  it("allows wildcard-allowlisted hostnames in strict mode", async () => {
+    const lookupFn = createLookupFn("93.184.216.34");
+    await expect(
+      assertBrowserNavigationAllowed({
+        url: "https://sub.example.com",
+        lookupFn,
+        ssrfPolicy: {
+          dangerouslyAllowPrivateNetwork: false,
+          hostnameAllowlist: ["*.example.com"],
+        },
+      }),
+    ).resolves.toBeUndefined();
+  });
+
+  it("does not treat the bare suffix as matching a wildcard allowlist entry", async () => {
+    const lookupFn = createLookupFn("93.184.216.34");
+    await expect(
+      assertBrowserNavigationAllowed({
+        url: "https://example.com",
+        lookupFn,
+        ssrfPolicy: {
+          dangerouslyAllowPrivateNetwork: false,
+          hostnameAllowlist: ["*.example.com"],
+        },
+      }),
+    ).rejects.toThrow(/dns rebinding protections are unavailable/i);
+    expect(lookupFn).not.toHaveBeenCalled();
+  });
+
+  it("does not match sibling domains against wildcard allowlist entries", async () => {
+    const lookupFn = createLookupFn("93.184.216.34");
+    await expect(
+      assertBrowserNavigationAllowed({
+        url: "https://evil-example.com",
+        lookupFn,
+        ssrfPolicy: {
+          dangerouslyAllowPrivateNetwork: false,
+          hostnameAllowlist: ["*.example.com"],
+        },
+      }),
+    ).rejects.toThrow(/dns rebinding protections are unavailable/i);
+    expect(lookupFn).not.toHaveBeenCalled();
+  });
+
+  it("treats bracketed IPv6 URL hostnames as IP literals in strict mode", async () => {
+    await expect(
+      assertBrowserNavigationAllowed({
+        url: "https://[2606:4700:4700::1111]/",
+        ssrfPolicy: { dangerouslyAllowPrivateNetwork: false },
+      }),
+    ).resolves.toBeUndefined();
+  });
+
   it("blocks strict policy navigation when env proxy is configured", async () => {
     vi.stubEnv("HTTP_PROXY", "http://127.0.0.1:7890");
     const lookupFn = createLookupFn("93.184.216.34");
@@ -165,6 +244,15 @@ describe("browser navigation guard", () => {
     ).resolves.toBeUndefined();
   });
 
+  it("blocks final hostname URLs in strict mode after navigation", async () => {
+    await expect(
+      assertBrowserNavigationResultAllowed({
+        url: "https://example.com/final",
+        ssrfPolicy: { dangerouslyAllowPrivateNetwork: false },
+      }),
+    ).rejects.toBeInstanceOf(InvalidBrowserNavigationUrlError);
+  });
+
   it("blocks private intermediate redirect hops", async () => {
     const publicLookup = createLookupFn("93.184.216.34");
     const privateLookup = createLookupFn("127.0.0.1");

extensions/browser/src/browser/navigation-guard.ts

@@ -1,3 +1,8 @@
+import { isIP } from "node:net";
+import {
+  matchesHostnameAllowlist,
+  normalizeHostname,
+} from "openclaw/plugin-sdk/browser-security-runtime";
 import { normalizeOptionalString } from "openclaw/plugin-sdk/text-runtime";
 import { hasProxyEnvConfigured } from "../infra/net/proxy-env.js";
 import {
@@ -41,6 +46,24 @@ export function requiresInspectableBrowserNavigationRedirects(ssrfPolicy?: SsrFP
   return !isPrivateNetworkAllowedByPolicy(ssrfPolicy);
 }
 
+function isIpLiteralHostname(hostname: string): boolean {
+  return isIP(normalizeHostname(hostname)) !== 0;
+}
+
+function isExplicitlyAllowedBrowserHostname(hostname: string, ssrfPolicy?: SsrFPolicy): boolean {
+  const normalizedHostname = normalizeHostname(hostname);
+  const exactMatches = ssrfPolicy?.allowedHostnames ?? [];
+  if (exactMatches.some((value) => normalizeHostname(value) === normalizedHostname)) {
+    return true;
+  }
+  const hostnameAllowlist = (ssrfPolicy?.hostnameAllowlist ?? [])
+    .map((pattern) => normalizeHostname(pattern))
+    .filter(Boolean);
+  return hostnameAllowlist.length > 0
+    ? matchesHostnameAllowlist(normalizedHostname, hostnameAllowlist)
+    : false;
+}
+
 export async function assertBrowserNavigationAllowed(
   opts: {
     url: string;
@@ -78,6 +101,21 @@ export async function assertBrowserNavigationAllowed(
     );
   }
 
+  // Browser navigations happen in Chromium's network stack, not Node's. In
+  // strict mode, a hostname-based URL would be resolved twice by different
+  // resolvers, so Node-side pinning cannot guarantee the browser connects to
+  // the same address that passed policy checks.
+  if (
+    opts.ssrfPolicy &&
+    !isPrivateNetworkAllowedByPolicy(opts.ssrfPolicy) &&
+    !isIpLiteralHostname(parsed.hostname) &&
+    !isExplicitlyAllowedBrowserHostname(parsed.hostname, opts.ssrfPolicy)
+  ) {
+    throw new InvalidBrowserNavigationUrlError(
+      "Navigation blocked: strict browser SSRF policy requires an IP-literal URL because browser DNS rebinding protections are unavailable for hostname-based navigation",
+    );
+  }
+
   await resolvePinnedHostnameWithPolicy(parsed.hostname, {
     lookupFn: opts.lookupFn,
     policy: opts.ssrfPolicy,
@@ -87,7 +125,8 @@ export async function assertBrowserNavigationAllowed(
 /**
  * Best-effort post-navigation guard for final page URLs.
  * Only validates network URLs (http/https) and about:blank to avoid false
- * positives on browser-internal error pages (e.g. chrome-error://).
+ * positives on browser-internal error pages (e.g. chrome-error://). In strict
+ * mode this intentionally re-applies the hostname gate after redirects.
  */
 export async function assertBrowserNavigationResultAllowed(
   opts: {

extensions/browser/src/browser/profiles-service.test.ts

@@ -111,7 +111,9 @@ describe("BrowserProfilesService", () => {
   });
 
   it("accepts per-profile cdpUrl for remote Chrome", async () => {
-    const resolved = resolveBrowserConfig({});
+    const resolved = resolveBrowserConfig({
+      ssrfPolicy: { dangerouslyAllowPrivateNetwork: true },
+    });
     const { ctx } = createCtx(resolved);
 
     vi.mocked(loadConfig).mockReturnValue({ browser: { profiles: {} } });

extensions/browser/src/browser/profiles-service.ts

@@ -124,6 +124,11 @@ export function createBrowserProfilesService(ctx: BrowserRouteContext) {
     }
 
     if (rawCdpUrl) {
+      if (driver === "existing-session") {
+        throw new BrowserValidationError(
+          "driver=existing-session does not accept cdpUrl; it attaches via the Chrome MCP auto-connect flow",
+        );
+      }
       let parsed: ReturnType<typeof parseHttpUrl>;
       try {
         parsed = parseHttpUrl(rawCdpUrl, "browser.profiles.cdpUrl");
@@ -131,11 +136,6 @@ export function createBrowserProfilesService(ctx: BrowserRouteContext) {
       } catch (err) {
         throw new BrowserValidationError(formatErrorMessage(err));
       }
-      if (driver === "existing-session") {
-        throw new BrowserValidationError(
-          "driver=existing-session does not accept cdpUrl; it attaches via the Chrome MCP auto-connect flow",
-        );
-      }
       profileConfig = {
         cdpUrl: parsed.normalized,
         ...(driver ? { driver } : {}),

extensions/browser/src/browser/pw-role-snapshot.ts

@@ -340,7 +340,7 @@ export function buildRoleSnapshotFromAiSnapshot(
   aiSnapshot: string,
   options: RoleSnapshotOptions = {},
 ): { snapshot: string; refs: RoleRefMap } {
-  const lines = String(aiSnapshot ?? "").split("\n");
+  const lines = aiSnapshot.split("\n");
   const refs: RoleRefMap = {};
 
   if (options.interactive) {

extensions/browser/src/browser/pw-session.create-page.navigation-guard.test.ts

@@ -1,5 +1,5 @@
 import { chromium } from "playwright-core";
-import { afterEach, describe, expect, it, vi } from "vitest";
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
 import { SsrFBlockedError } from "../infra/net/ssrf.js";
 import * as chromeModule from "./chrome.js";
 import { BrowserTabNotFoundError } from "./errors.js";
@@ -15,9 +15,33 @@ import {
   listPagesViaPlaywright,
 } from "./pw-session.js";
 
+vi.mock("openclaw/plugin-sdk/browser-security-runtime", async () => {
+  const actual = await vi.importActual<
+    typeof import("openclaw/plugin-sdk/browser-security-runtime")
+  >("openclaw/plugin-sdk/browser-security-runtime");
+  const lookupFn = async (_hostname: string, options?: { all?: boolean }) => {
+    const result = { address: "93.184.216.34", family: 4 };
+    return options?.all === true ? [result] : result;
+  };
+  return {
+    ...actual,
+    resolvePinnedHostnameWithPolicy: (hostname: string, params: object = {}) =>
+      actual.resolvePinnedHostnameWithPolicy(hostname, { ...params, lookupFn: lookupFn as never }),
+  };
+});
+
 const connectOverCdpSpy = vi.spyOn(chromium, "connectOverCDP");
 const getChromeWebSocketUrlSpy = vi.spyOn(chromeModule, "getChromeWebSocketUrl");
 
+const PROXY_ENV_KEYS = [
+  "ALL_PROXY",
+  "all_proxy",
+  "HTTP_PROXY",
+  "http_proxy",
+  "HTTPS_PROXY",
+  "https_proxy",
+] as const;
+
 type MockRoute = { continue: () => Promise<void>; abort: () => Promise<void> };
 type MockRequest = {
   isNavigationRequest: () => boolean;
@@ -126,6 +150,7 @@ async function dispatchMockNavigation(params: {
   getRouteHandler: () => MockRouteHandler | null;
   mainFrame: object;
   url: string;
+  frame?: object;
   isNavigationRequest?: boolean;
   resourceType?: string;
   route?: Partial<MockRoute>;
@@ -137,7 +162,7 @@ async function dispatchMockNavigation(params: {
   const { resourceType } = params;
   await handler(createMockRoute(params.route), {
     isNavigationRequest: () => params.isNavigationRequest ?? true,
-    frame: () => params.mainFrame,
+    frame: () => params.frame ?? params.mainFrame,
     ...(resourceType ? { resourceType: () => resourceType } : {}),
     url: () => params.url,
   });
@@ -169,7 +194,14 @@ function mockBlockedRedirectNavigation(params: {
   });
 }
 
+beforeEach(() => {
+  for (const key of PROXY_ENV_KEYS) {
+    vi.stubEnv(key, "");
+  }
+});
+
 afterEach(async () => {
+  vi.unstubAllEnvs();
   connectOverCdpSpy.mockClear();
   getChromeWebSocketUrlSpy.mockClear();
   await closePlaywrightBrowserConnection().catch(() => {});
@@ -201,6 +233,20 @@ describe("pw-session createPageViaPlaywright navigation guard", () => {
     expect(pageGoto).not.toHaveBeenCalled();
   });
 
+  it("blocks hostname navigation when strict SSRF policy is configured", async () => {
+    const { pageGoto } = installBrowserMocks();
+
+    await expect(
+      createPageViaPlaywright({
+        cdpUrl: "http://127.0.0.1:18792",
+        url: "https://example.com",
+        ssrfPolicy: { dangerouslyAllowPrivateNetwork: false, allowedHostnames: ["127.0.0.1"] },
+      }),
+    ).rejects.toBeInstanceOf(InvalidBrowserNavigationUrlError);
+
+    expect(pageGoto).not.toHaveBeenCalled();
+  });
+
   it("blocks private intermediate redirect hops", async () => {
     const { pageGoto, pageClose, getRouteHandler, mainFrame } = installBrowserMocks();
     mockBlockedRedirectNavigation({ pageGoto, getRouteHandler, mainFrame });
@@ -237,6 +283,41 @@ describe("pw-session createPageViaPlaywright navigation guard", () => {
     expect(pageClose).toHaveBeenCalledTimes(1);
   });
 
+  it("aborts private subframe document hops without quarantining the page", async () => {
+    const { pageGoto, pageClose, getRouteHandler, mainFrame } = installBrowserMocks();
+    const subframe = {};
+    const subframeRoute = createMockRoute();
+    pageGoto.mockImplementationOnce(async () => {
+      await dispatchMockNavigation({
+        getRouteHandler,
+        mainFrame,
+        url: "https://93.184.216.34/start",
+      });
+      await dispatchMockNavigation({
+        getRouteHandler,
+        mainFrame,
+        frame: subframe,
+        url: "http://127.0.0.1:18080/internal-hop",
+        route: subframeRoute,
+      });
+      return {
+        request: () => ({
+          url: () => "https://93.184.216.34/start",
+          redirectedFrom: () => null,
+        }),
+      };
+    });
+
+    const created = await createPageViaPlaywright({
+      cdpUrl: "http://127.0.0.1:18792",
+      url: "https://93.184.216.34/start",
+    });
+
+    expect(created.targetId).toBe("TARGET_1");
+    expect(subframeRoute.abort).toHaveBeenCalledTimes(1);
+    expect(pageClose).not.toHaveBeenCalled();
+  });
+
   it("preserves the created tab on ordinary navigation failure", async () => {
     const { pageGoto, pageClose } = installBrowserMocks();
     pageGoto.mockRejectedValueOnce(new Error("page.goto: net::ERR_NAME_NOT_RESOLVED"));

extensions/browser/src/browser/pw-session.ts

@@ -14,6 +14,7 @@ import { SsrFBlockedError, type SsrFPolicy } from "../infra/net/ssrf.js";
 import { withNoProxyForCdpUrl } from "./cdp-proxy-bypass.js";
 import {
   appendCdpPath,
+  assertCdpEndpointAllowed,
   fetchJson,
   getHeadersWithAuth,
   normalizeCdpHttpBaseForJsonEndpoints,
@@ -337,9 +338,9 @@ export function ensurePageState(page: Page): PageState {
     });
     page.on("pageerror", (err: Error) => {
       state.errors.push({
-        message: err?.message ? String(err.message) : String(err),
-        name: err?.name ? String(err.name) : undefined,
-        stack: err?.stack ? String(err.stack) : undefined,
+        message: err.message || String(err),
+        name: err.name || undefined,
+        stack: err.stack || undefined,
         timestamp: new Date().toISOString(),
       });
       if (state.errors.length > MAX_PAGE_ERRORS) {
@@ -424,12 +425,15 @@ function observeBrowser(browser: Browser) {
   }
 }
 
-async function connectBrowser(cdpUrl: string): Promise<ConnectedBrowser> {
+async function connectBrowser(cdpUrl: string, ssrfPolicy?: SsrFPolicy): Promise<ConnectedBrowser> {
   const normalized = normalizeCdpUrl(cdpUrl);
   const cached = cachedByCdpUrl.get(normalized);
   if (cached) {
     return cached;
   }
+  // Run SSRF policy check only on cache miss so transient DNS failures
+  // do not break active sessions that already hold a live CDP connection.
+  await assertCdpEndpointAllowed(normalized, ssrfPolicy);
   const connecting = connectingByCdpUrl.get(normalized);
   if (connecting) {
     return await connecting;
@@ -440,7 +444,9 @@ async function connectBrowser(cdpUrl: string): Promise<ConnectedBrowser> {
     for (let attempt = 0; attempt < 3; attempt += 1) {
       try {
         const timeout = 5000 + attempt * 2000;
-        const wsUrl = await getChromeWebSocketUrl(normalized, timeout).catch(() => null);
+        const wsUrl = await getChromeWebSocketUrl(normalized, timeout, ssrfPolicy).catch(
+          () => null,
+        );
         const endpoint = wsUrl ?? normalized;
         const headers = getHeadersWithAuth(endpoint);
         // Bypass proxy for loopback CDP connections (#31219)
@@ -562,8 +568,10 @@ async function findPageByTargetIdViaTargetList(
   pages: Page[],
   targetId: string,
   cdpUrl: string,
+  ssrfPolicy?: SsrFPolicy,
 ): Promise<Page | null> {
   const cdpHttpBase = normalizeCdpHttpBaseForJsonEndpoints(cdpUrl);
+  await assertCdpEndpointAllowed(cdpUrl, ssrfPolicy);
   const targets = await fetchJson<
     Array<{
       id: string;
@@ -578,6 +586,7 @@ async function findPageByTargetId(
   browser: Browser,
   targetId: string,
   cdpUrl?: string,
+  ssrfPolicy?: SsrFPolicy,
 ): Promise<Page | null> {
   const pages = await getAllPages(browser);
   let resolvedViaCdp = false;
@@ -595,7 +604,7 @@ async function findPageByTargetId(
   }
   if (cdpUrl) {
     try {
-      return await findPageByTargetIdViaTargetList(pages, targetId, cdpUrl);
+      return await findPageByTargetIdViaTargetList(pages, targetId, cdpUrl, ssrfPolicy);
     } catch {
       // Ignore fetch errors and fall through to return null.
     }
@@ -609,12 +618,13 @@ async function findPageByTargetId(
 async function resolvePageByTargetIdOrThrow(opts: {
   cdpUrl: string;
   targetId: string;
+  ssrfPolicy?: SsrFPolicy;
 }): Promise<Page> {
   if (isBlockedTarget(opts.cdpUrl, opts.targetId)) {
     throw new BlockedBrowserTargetError();
   }
-  const { browser } = await connectBrowser(opts.cdpUrl);
-  const page = await findPageByTargetId(browser, opts.targetId, opts.cdpUrl);
+  const { browser } = await connectBrowser(opts.cdpUrl, opts.ssrfPolicy);
+  const page = await findPageByTargetId(browser, opts.targetId, opts.cdpUrl, opts.ssrfPolicy);
   if (!page) {
     throw new BrowserTabNotFoundError();
   }
@@ -624,11 +634,12 @@ async function resolvePageByTargetIdOrThrow(opts: {
 export async function getPageForTargetId(opts: {
   cdpUrl: string;
   targetId?: string;
+  ssrfPolicy?: SsrFPolicy;
 }): Promise<Page> {
   if (opts.targetId && isBlockedTarget(opts.cdpUrl, opts.targetId)) {
     throw new BlockedBrowserTargetError();
   }
-  const { browser } = await connectBrowser(opts.cdpUrl);
+  const { browser } = await connectBrowser(opts.cdpUrl, opts.ssrfPolicy);
   const pages = await getAllPages(browser);
   if (!pages.length) {
     throw new Error("No pages available in the connected browser.");
@@ -648,7 +659,7 @@ export async function getPageForTargetId(opts: {
   if (!opts.targetId) {
     return first;
   }
-  const found = await findPageByTargetId(browser, opts.targetId, opts.cdpUrl);
+  const found = await findPageByTargetId(browser, opts.targetId, opts.cdpUrl, opts.ssrfPolicy);
   if (found) {
     if (isBlockedPageRef(opts.cdpUrl, found)) {
       throw new BlockedBrowserTargetError();
@@ -693,6 +704,36 @@ function isTopLevelNavigationRequest(page: Page, request: Request): boolean {
   }
 }
 
+function isSubframeDocumentNavigationRequest(page: Page, request: Request): boolean {
+  let sameMainFrame = false;
+  try {
+    sameMainFrame = request.frame() === page.mainFrame();
+  } catch {
+    // Fail closed: if frame resolution throws after the top-level check already
+    // determined this is NOT the main frame, treat it as a subframe document
+    // navigation so the SSRF guard still fires. Returning false here would let
+    // transient renderer churn skip the policy check entirely.
+    return true;
+  }
+  if (sameMainFrame) {
+    return false;
+  }
+
+  try {
+    if (request.isNavigationRequest()) {
+      return true;
+    }
+  } catch {
+    // Fall through to the resource-type check.
+  }
+
+  try {
+    return request.resourceType() === "document";
+  } catch {
+    return false;
+  }
+}
+
 function isPolicyDenyNavigationError(err: unknown): boolean {
   return err instanceof SsrFBlockedError || err instanceof InvalidBrowserNavigationUrlError;
 }
@@ -758,7 +799,10 @@ export async function gotoPageWithNavigationGuard(opts: {
       await route.abort().catch(() => {});
       return;
     }
-    if (!isTopLevelNavigationRequest(opts.page, request)) {
+    const isTopLevel = isTopLevelNavigationRequest(opts.page, request);
+    const isSubframeDocument =
+      !isTopLevel && isSubframeDocumentNavigationRequest(opts.page, request);
+    if (!isTopLevel && !isSubframeDocument) {
       await route.continue();
       return;
     }
@@ -769,7 +813,9 @@ export async function gotoPageWithNavigationGuard(opts: {
       });
     } catch (err) {
       if (isPolicyDenyNavigationError(err)) {
-        blockedError = err;
+        if (isTopLevel) {
+          blockedError = err;
+        }
         await route.abort().catch(() => {});
         return;
       }
@@ -887,7 +933,9 @@ function cdpSocketNeedsAttach(wsUrl: string): boolean {
 async function tryTerminateExecutionViaCdp(opts: {
   cdpUrl: string;
   targetId: string;
+  ssrfPolicy?: SsrFPolicy;
 }): Promise<void> {
+  await assertCdpEndpointAllowed(opts.cdpUrl, opts.ssrfPolicy);
   const cdpHttpBase = normalizeCdpHttpBaseForJsonEndpoints(opts.cdpUrl);
   const listUrl = appendCdpPath(cdpHttpBase, "/json/list");
 
@@ -976,6 +1024,7 @@ export async function forceDisconnectPlaywrightForTarget(opts: {
   cdpUrl: string;
   targetId?: string;
   reason?: string;
+  ssrfPolicy?: SsrFPolicy;
 }): Promise<void> {
   const normalized = normalizeCdpUrl(opts.cdpUrl);
   const cur = cachedByCdpUrl.get(normalized);
@@ -996,7 +1045,11 @@ export async function forceDisconnectPlaywrightForTarget(opts: {
   // disconnect Playwright's CDP connection.
   const targetId = normalizeOptionalString(opts.targetId) ?? "";
   if (targetId) {
-    await tryTerminateExecutionViaCdp({ cdpUrl: normalized, targetId }).catch(() => {});
+    await tryTerminateExecutionViaCdp({
+      cdpUrl: normalized,
+      targetId,
+      ssrfPolicy: opts.ssrfPolicy,
+    }).catch(() => {});
   }
 
   // Fire-and-forget: don't await because browser.close() may hang on the stuck CDP pipe.
@@ -1007,7 +1060,10 @@ export async function forceDisconnectPlaywrightForTarget(opts: {
  * List all pages/tabs from the persistent Playwright connection.
  * Used for remote profiles where HTTP-based /json/list is ephemeral.
  */
-export async function listPagesViaPlaywright(opts: { cdpUrl: string }): Promise<
+export async function listPagesViaPlaywright(opts: {
+  cdpUrl: string;
+  ssrfPolicy?: SsrFPolicy;
+}): Promise<
   Array<{
     targetId: string;
     title: string;
@@ -1015,7 +1071,7 @@ export async function listPagesViaPlaywright(opts: { cdpUrl: string }): Promise<
     type: string;
   }>
 > {
-  const { browser } = await connectBrowser(opts.cdpUrl);
+  const { browser } = await connectBrowser(opts.cdpUrl, opts.ssrfPolicy);
   const pages = await getAllPages(browser);
   const results: Array<{
     targetId: string;
@@ -1056,7 +1112,7 @@ export async function createPageViaPlaywright(opts: {
   url: string;
   type: string;
 }> {
-  const { browser } = await connectBrowser(opts.cdpUrl);
+  const { browser } = await connectBrowser(opts.cdpUrl, opts.ssrfPolicy);
   const context = browser.contexts()[0] ?? (await browser.newContext());
   ensureContextState(context);
 
@@ -1119,6 +1175,7 @@ export async function createPageViaPlaywright(opts: {
 export async function closePageByTargetIdViaPlaywright(opts: {
   cdpUrl: string;
   targetId: string;
+  ssrfPolicy?: SsrFPolicy;
 }): Promise<void> {
   const page = await resolvePageByTargetIdOrThrow(opts);
   await page.close();
@@ -1131,6 +1188,7 @@ export async function closePageByTargetIdViaPlaywright(opts: {
 export async function focusPageByTargetIdViaPlaywright(opts: {
   cdpUrl: string;
   targetId: string;
+  ssrfPolicy?: SsrFPolicy;
 }): Promise<void> {
   const page = await resolvePageByTargetIdOrThrow(opts);
   try {

extensions/browser/src/browser/pw-tools-core.downloads.ts

@@ -256,7 +256,7 @@ export async function downloadViaPlaywright(opts: {
   const timeout = normalizeTimeoutMs(opts.timeoutMs, 120_000);
 
   const ref = requireRef(opts.ref);
-  const outPath = String(opts.path ?? "").trim();
+  const outPath = opts.path?.trim() ?? "";
   if (!outPath) {
     throw new Error("path is required");
   }

extensions/browser/src/browser/pw-tools-core.interactions.navigation-guard.test.ts

@@ -1,5 +1,6 @@
 import { describe, expect, it, vi } from "vitest";
 import {
+  getPwToolsCoreNavigationGuardMocks,
   getPwToolsCoreSessionMocks,
   installPwToolsCoreTestHooks,
   setPwToolsCoreCurrentPage,
@@ -9,6 +10,18 @@ import {
 installPwToolsCoreTestHooks();
 const mod = await import("./pw-tools-core.js");
 
+function createMutableFrame(initialUrl: string) {
+  let currentUrl = initialUrl;
+  return {
+    frame: {
+      url: vi.fn(() => currentUrl),
+    },
+    setUrl: (nextUrl: string) => {
+      currentUrl = nextUrl;
+    },
+  };
+}
+
 describe("pw-tools-core interaction navigation guard", () => {
   it("waits for the grace window before completing a successful non-navigating click", async () => {
     vi.useFakeTimers();
@@ -120,12 +133,12 @@ describe("pw-tools-core interaction navigation guard", () => {
     }
   });
 
-  it("ignores subframe framenavigated events before the main frame navigates", async () => {
+  it("checks subframe navigations before a later main-frame navigation", async () => {
     vi.useFakeTimers();
     try {
       const listeners = new Set<(frame: object) => void>();
       const mainFrame = {};
-      const subframe = {};
+      const subframe = { url: () => "https://example.com/embed" };
       let currentUrl = "http://127.0.0.1:9222/json/version";
       const click = vi.fn(async () => {
         setTimeout(() => {
@@ -169,10 +182,449 @@ describe("pw-tools-core interaction navigation guard", () => {
       expect(
         getPwToolsCoreSessionMocks().assertPageNavigationCompletedSafely,
       ).not.toHaveBeenCalled();
+      expect(
+        getPwToolsCoreNavigationGuardMocks().assertBrowserNavigationResultAllowed,
+      ).not.toHaveBeenCalled();
 
       await vi.advanceTimersByTimeAsync(10);
       await task;
 
+      expect(
+        getPwToolsCoreNavigationGuardMocks().assertBrowserNavigationResultAllowed,
+      ).toHaveBeenCalledWith({
+        ssrfPolicy: { allowPrivateNetwork: false },
+        url: "https://example.com/embed",
+      });
+      expect(getPwToolsCoreSessionMocks().assertPageNavigationCompletedSafely).toHaveBeenCalledWith(
+        {
+          cdpUrl: "http://127.0.0.1:18792",
+          page,
+          response: null,
+          ssrfPolicy: { allowPrivateNetwork: false },
+          targetId: "T1",
+        },
+      );
+    } finally {
+      vi.useRealTimers();
+    }
+  });
+
+  it("blocks subframe-only navigation to a private URL during the post-action grace window", async () => {
+    vi.useFakeTimers();
+    try {
+      const listeners = new Set<(frame: object) => void>();
+      const mainFrame = {};
+      const subframe = { url: () => "http://169.254.169.254/latest/meta-data/" };
+      const click = vi.fn(async () => {
+        setTimeout(() => {
+          for (const listener of listeners) {
+            listener(subframe);
+          }
+        }, 10);
+      });
+      const page = {
+        mainFrame: vi.fn(() => mainFrame),
+        on: vi.fn((event: string, listener: (frame: object) => void) => {
+          if (event === "framenavigated") {
+            listeners.add(listener);
+          }
+        }),
+        off: vi.fn((event: string, listener: (frame: object) => void) => {
+          if (event === "framenavigated") {
+            listeners.delete(listener);
+          }
+        }),
+        url: vi.fn(() => "https://attacker.example.com/page"),
+      };
+      setPwToolsCoreCurrentRefLocator({ click });
+      setPwToolsCoreCurrentPage(page);
+
+      const blocked = new Error("SSRF blocked: private network");
+      getPwToolsCoreNavigationGuardMocks().assertBrowserNavigationResultAllowed.mockRejectedValueOnce(
+        blocked,
+      );
+
+      const task = mod.clickViaPlaywright({
+        cdpUrl: "http://127.0.0.1:18792",
+        targetId: "T1",
+        ref: "1",
+        ssrfPolicy: { allowPrivateNetwork: false },
+      });
+      const rejection = expect(task).rejects.toThrow("SSRF blocked: private network");
+
+      await vi.advanceTimersByTimeAsync(10);
+      await vi.advanceTimersByTimeAsync(240);
+      await rejection;
+      expect(
+        getPwToolsCoreSessionMocks().assertPageNavigationCompletedSafely,
+      ).not.toHaveBeenCalled();
+    } finally {
+      vi.useRealTimers();
+    }
+  });
+
+  it("snapshots delayed subframe URLs before later rewrites make them look safe", async () => {
+    vi.useFakeTimers();
+    try {
+      const listeners = new Set<(frame: object) => void>();
+      const mainFrame = {};
+      const subframe = createMutableFrame("http://169.254.169.254/latest/meta-data/");
+      const click = vi.fn(async () => {
+        setTimeout(() => {
+          for (const listener of listeners) {
+            listener(subframe.frame);
+          }
+        }, 10);
+        setTimeout(() => {
+          subframe.setUrl("https://example.com/embed");
+        }, 20);
+      });
+      const page = {
+        mainFrame: vi.fn(() => mainFrame),
+        on: vi.fn((event: string, listener: (frame: object) => void) => {
+          if (event === "framenavigated") {
+            listeners.add(listener);
+          }
+        }),
+        off: vi.fn((event: string, listener: (frame: object) => void) => {
+          if (event === "framenavigated") {
+            listeners.delete(listener);
+          }
+        }),
+        url: vi.fn(() => "https://attacker.example.com/page"),
+      };
+      setPwToolsCoreCurrentRefLocator({ click });
+      setPwToolsCoreCurrentPage(page);
+
+      const task = mod.clickViaPlaywright({
+        cdpUrl: "http://127.0.0.1:18792",
+        targetId: "T1",
+        ref: "1",
+        ssrfPolicy: { allowPrivateNetwork: false },
+      });
+
+      await vi.advanceTimersByTimeAsync(20);
+      await vi.advanceTimersByTimeAsync(230);
+      await task;
+
+      expect(
+        getPwToolsCoreNavigationGuardMocks().assertBrowserNavigationResultAllowed,
+      ).toHaveBeenCalledWith({
+        ssrfPolicy: { allowPrivateNetwork: false },
+        url: "http://169.254.169.254/latest/meta-data/",
+      });
+    } finally {
+      vi.useRealTimers();
+    }
+  });
+
+  it("still quarantines the main frame when a delayed subframe block fires first", async () => {
+    vi.useFakeTimers();
+    try {
+      const listeners = new Set<(frame: object) => void>();
+      const mainFrame = {};
+      const subframe = { url: () => "http://169.254.169.254/latest/meta-data/" };
+      let currentUrl = "https://attacker.example.com/page";
+      const click = vi.fn(async () => {
+        setTimeout(() => {
+          for (const listener of listeners) {
+            listener(subframe);
+          }
+        }, 10);
+        setTimeout(() => {
+          currentUrl = "http://127.0.0.1:8080/internal";
+          for (const listener of listeners) {
+            listener(mainFrame);
+          }
+        }, 20);
+      });
+      const page = {
+        mainFrame: vi.fn(() => mainFrame),
+        on: vi.fn((event: string, listener: (frame: object) => void) => {
+          if (event === "framenavigated") {
+            listeners.add(listener);
+          }
+        }),
+        off: vi.fn((event: string, listener: (frame: object) => void) => {
+          if (event === "framenavigated") {
+            listeners.delete(listener);
+          }
+        }),
+        url: vi.fn(() => currentUrl),
+      };
+      setPwToolsCoreCurrentRefLocator({ click });
+      setPwToolsCoreCurrentPage(page);
+
+      const subframeBlocked = new Error("subframe blocked");
+      const mainFrameBlocked = new Error("main frame blocked");
+      getPwToolsCoreNavigationGuardMocks().assertBrowserNavigationResultAllowed.mockRejectedValueOnce(
+        subframeBlocked,
+      );
+      getPwToolsCoreSessionMocks().assertPageNavigationCompletedSafely.mockRejectedValueOnce(
+        mainFrameBlocked,
+      );
+
+      const task = mod.clickViaPlaywright({
+        cdpUrl: "http://127.0.0.1:18792",
+        targetId: "T1",
+        ref: "1",
+        ssrfPolicy: { allowPrivateNetwork: false },
+      });
+      const rejection = expect(task).rejects.toThrow("main frame blocked");
+
+      await vi.advanceTimersByTimeAsync(20);
+      await rejection;
+      expect(getPwToolsCoreSessionMocks().assertPageNavigationCompletedSafely).toHaveBeenCalledWith(
+        {
+          cdpUrl: "http://127.0.0.1:18792",
+          page,
+          response: null,
+          ssrfPolicy: { allowPrivateNetwork: false },
+          targetId: "T1",
+        },
+      );
+    } finally {
+      vi.useRealTimers();
+    }
+  });
+
+  it("does not stop watching for a later main-frame navigation after a harmless subframe hop", async () => {
+    vi.useFakeTimers();
+    try {
+      const listeners = new Set<(frame: object) => void>();
+      const mainFrame = {};
+      const subframe = { url: () => "about:blank" };
+      let currentUrl = "http://127.0.0.1:9222/json/version";
+      const click = vi.fn(async () => {
+        setTimeout(() => {
+          for (const listener of listeners) {
+            listener(subframe);
+          }
+        }, 10);
+        setTimeout(() => {
+          currentUrl = "http://127.0.0.1:9222/json/list";
+          for (const listener of listeners) {
+            listener(mainFrame);
+          }
+        }, 20);
+      });
+      const page = {
+        mainFrame: vi.fn(() => mainFrame),
+        on: vi.fn((event: string, listener: (frame: object) => void) => {
+          if (event === "framenavigated") {
+            listeners.add(listener);
+          }
+        }),
+        off: vi.fn((event: string, listener: (frame: object) => void) => {
+          if (event === "framenavigated") {
+            listeners.delete(listener);
+          }
+        }),
+        url: vi.fn(() => currentUrl),
+      };
+      setPwToolsCoreCurrentRefLocator({ click });
+      setPwToolsCoreCurrentPage(page);
+
+      const task = mod.clickViaPlaywright({
+        cdpUrl: "http://127.0.0.1:18792",
+        targetId: "T1",
+        ref: "1",
+        ssrfPolicy: { allowPrivateNetwork: false },
+      });
+
+      await vi.advanceTimersByTimeAsync(20);
+      await task;
+
+      expect(
+        getPwToolsCoreNavigationGuardMocks().assertBrowserNavigationResultAllowed,
+      ).not.toHaveBeenCalled();
+      expect(getPwToolsCoreSessionMocks().assertPageNavigationCompletedSafely).toHaveBeenCalledWith(
+        {
+          cdpUrl: "http://127.0.0.1:18792",
+          page,
+          response: null,
+          ssrfPolicy: { allowPrivateNetwork: false },
+          targetId: "T1",
+        },
+      );
+    } finally {
+      vi.useRealTimers();
+    }
+  });
+
+  it("checks delayed subframe navigations in the action-error recovery path", async () => {
+    vi.useFakeTimers();
+    try {
+      const listeners = new Set<(frame: object) => void>();
+      const mainFrame = {};
+      const subframe = { url: () => "http://169.254.169.254/latest/meta-data/" };
+      const page = {
+        mainFrame: vi.fn(() => mainFrame),
+        evaluate: vi.fn(async () => {
+          setTimeout(() => {
+            for (const listener of listeners) {
+              listener(subframe);
+            }
+          }, 10);
+          throw new Error("evaluate failed");
+        }),
+        on: vi.fn((event: string, listener: (frame: object) => void) => {
+          if (event === "framenavigated") {
+            listeners.add(listener);
+          }
+        }),
+        off: vi.fn((event: string, listener: (frame: object) => void) => {
+          if (event === "framenavigated") {
+            listeners.delete(listener);
+          }
+        }),
+        url: vi.fn(() => "https://attacker.example.com/page"),
+      };
+      setPwToolsCoreCurrentPage(page);
+
+      const blocked = new Error("SSRF blocked: private network");
+      getPwToolsCoreNavigationGuardMocks().assertBrowserNavigationResultAllowed.mockRejectedValueOnce(
+        blocked,
+      );
+
+      const task = mod.evaluateViaPlaywright({
+        cdpUrl: "http://127.0.0.1:18792",
+        targetId: "T1",
+        fn: "() => 1",
+        ssrfPolicy: { allowPrivateNetwork: false },
+      });
+      const rejection = expect(task).rejects.toThrow("SSRF blocked: private network");
+
+      await vi.advanceTimersByTimeAsync(10);
+      await vi.advanceTimersByTimeAsync(240);
+      await rejection;
+      expect(
+        getPwToolsCoreSessionMocks().assertPageNavigationCompletedSafely,
+      ).not.toHaveBeenCalled();
+    } finally {
+      vi.useRealTimers();
+    }
+  });
+
+  it("snapshots subframe URLs observed during the action before they change", async () => {
+    vi.useFakeTimers();
+    try {
+      const listeners = new Set<(frame: object) => void>();
+      const mainFrame = {};
+      const subframe = createMutableFrame("http://169.254.169.254/latest/meta-data/");
+      const click = vi.fn(
+        () =>
+          new Promise<void>((resolve) => {
+            setTimeout(() => {
+              for (const listener of listeners) {
+                listener(subframe.frame);
+              }
+            }, 10);
+            setTimeout(() => {
+              subframe.setUrl("https://example.com/embed");
+            }, 20);
+            setTimeout(resolve, 30);
+          }),
+      );
+      const page = {
+        mainFrame: vi.fn(() => mainFrame),
+        on: vi.fn((event: string, listener: (frame: object) => void) => {
+          if (event === "framenavigated") {
+            listeners.add(listener);
+          }
+        }),
+        off: vi.fn((event: string, listener: (frame: object) => void) => {
+          if (event === "framenavigated") {
+            listeners.delete(listener);
+          }
+        }),
+        url: vi.fn(() => "https://attacker.example.com/page"),
+      };
+      setPwToolsCoreCurrentRefLocator({ click });
+      setPwToolsCoreCurrentPage(page);
+
+      const task = mod.clickViaPlaywright({
+        cdpUrl: "http://127.0.0.1:18792",
+        targetId: "T1",
+        ref: "1",
+        ssrfPolicy: { allowPrivateNetwork: false },
+      });
+
+      await vi.advanceTimersByTimeAsync(30);
+      await vi.advanceTimersByTimeAsync(250);
+      await task;
+
+      expect(
+        getPwToolsCoreNavigationGuardMocks().assertBrowserNavigationResultAllowed,
+      ).toHaveBeenCalledWith({
+        ssrfPolicy: { allowPrivateNetwork: false },
+        url: "http://169.254.169.254/latest/meta-data/",
+      });
+    } finally {
+      vi.useRealTimers();
+    }
+  });
+
+  it("still quarantines the main frame when an in-flight subframe block fires first", async () => {
+    vi.useFakeTimers();
+    try {
+      const listeners = new Set<(frame: object) => void>();
+      const mainFrame = {};
+      const subframe = { url: () => "http://169.254.169.254/latest/meta-data/" };
+      let currentUrl = "https://attacker.example.com/page";
+      const click = vi.fn(
+        () =>
+          new Promise<void>((resolve) => {
+            setTimeout(() => {
+              for (const listener of listeners) {
+                listener(subframe);
+              }
+            }, 10);
+            setTimeout(() => {
+              currentUrl = "http://127.0.0.1:8080/internal";
+              for (const listener of listeners) {
+                listener(mainFrame);
+              }
+            }, 20);
+            setTimeout(resolve, 30);
+          }),
+      );
+      const page = {
+        mainFrame: vi.fn(() => mainFrame),
+        on: vi.fn((event: string, listener: (frame: object) => void) => {
+          if (event === "framenavigated") {
+            listeners.add(listener);
+          }
+        }),
+        off: vi.fn((event: string, listener: (frame: object) => void) => {
+          if (event === "framenavigated") {
+            listeners.delete(listener);
+          }
+        }),
+        url: vi.fn(() => currentUrl),
+      };
+      setPwToolsCoreCurrentRefLocator({ click });
+      setPwToolsCoreCurrentPage(page);
+
+      const subframeBlocked = new Error("subframe blocked");
+      const mainFrameBlocked = new Error("main frame blocked");
+      getPwToolsCoreNavigationGuardMocks().assertBrowserNavigationResultAllowed.mockRejectedValueOnce(
+        subframeBlocked,
+      );
+      getPwToolsCoreSessionMocks().assertPageNavigationCompletedSafely.mockRejectedValueOnce(
+        mainFrameBlocked,
+      );
+
+      const task = mod.clickViaPlaywright({
+        cdpUrl: "http://127.0.0.1:18792",
+        targetId: "T1",
+        ref: "1",
+        ssrfPolicy: { allowPrivateNetwork: false },
+      });
+      const rejection = expect(task).rejects.toThrow("main frame blocked");
+
+      await vi.advanceTimersByTimeAsync(30);
+      await rejection;
       expect(getPwToolsCoreSessionMocks().assertPageNavigationCompletedSafely).toHaveBeenCalledWith(
         {
           cdpUrl: "http://127.0.0.1:18792",
@@ -398,6 +850,115 @@ describe("pw-tools-core interaction navigation guard", () => {
     });
   });
 
+  it("runs the post-keypress navigation guard when navigation starts shortly after the keypress resolves", async () => {
+    vi.useFakeTimers();
+    try {
+      const listeners = new Set<() => void>();
+      let currentUrl = "http://127.0.0.1:9222/json/version";
+      const page = {
+        keyboard: {
+          press: vi.fn(async () => {
+            setTimeout(() => {
+              currentUrl = "http://127.0.0.1:9222/private-target";
+              for (const listener of listeners) {
+                listener();
+              }
+            }, 10);
+          }),
+        },
+        on: vi.fn((event: string, listener: () => void) => {
+          if (event === "framenavigated") {
+            listeners.add(listener);
+          }
+        }),
+        off: vi.fn((event: string, listener: () => void) => {
+          if (event === "framenavigated") {
+            listeners.delete(listener);
+          }
+        }),
+        url: vi.fn(() => currentUrl),
+      };
+      setPwToolsCoreCurrentPage(page);
+
+      const task = mod.pressKeyViaPlaywright({
+        cdpUrl: "http://127.0.0.1:18792",
+        targetId: "T1",
+        key: "Enter",
+        ssrfPolicy: { allowPrivateNetwork: false },
+      });
+
+      await vi.advanceTimersByTimeAsync(10);
+      await task;
+
+      expect(getPwToolsCoreSessionMocks().assertPageNavigationCompletedSafely).toHaveBeenCalledWith(
+        {
+          cdpUrl: "http://127.0.0.1:18792",
+          page,
+          response: null,
+          ssrfPolicy: { allowPrivateNetwork: false },
+          targetId: "T1",
+        },
+      );
+    } finally {
+      vi.useRealTimers();
+    }
+  });
+
+  it("propagates blocked delayed submit navigation instead of reporting type success", async () => {
+    vi.useFakeTimers();
+    try {
+      const listeners = new Set<() => void>();
+      let currentUrl = "https://example.com/form";
+      const locator = {
+        fill: vi.fn(async () => {}),
+        press: vi.fn(async () => {
+          setTimeout(() => {
+            currentUrl = "http://127.0.0.1:9222/private-target";
+            for (const listener of listeners) {
+              listener();
+            }
+          }, 10);
+        }),
+      };
+      const page = {
+        on: vi.fn((event: string, listener: () => void) => {
+          if (event === "framenavigated") {
+            listeners.add(listener);
+          }
+        }),
+        off: vi.fn((event: string, listener: () => void) => {
+          if (event === "framenavigated") {
+            listeners.delete(listener);
+          }
+        }),
+        url: vi.fn(() => currentUrl),
+      };
+      setPwToolsCoreCurrentRefLocator(locator);
+      setPwToolsCoreCurrentPage(page);
+
+      const blocked = new Error("blocked delayed interaction navigation");
+      getPwToolsCoreSessionMocks().assertPageNavigationCompletedSafely.mockRejectedValueOnce(
+        blocked,
+      );
+
+      const task = mod.typeViaPlaywright({
+        cdpUrl: "http://127.0.0.1:18792",
+        targetId: "T1",
+        ref: "1",
+        text: "hello",
+        submit: true,
+        ssrfPolicy: { allowPrivateNetwork: false },
+      });
+      const rejection = expect(task).rejects.toThrow("blocked delayed interaction navigation");
+
+      await vi.advanceTimersByTimeAsync(10);
+      await rejection;
+      expect(listeners.size).toBe(0);
+    } finally {
+      vi.useRealTimers();
+    }
+  });
+
   it("does not run the post-click navigation guard when the url is unchanged", async () => {
     const click = vi.fn(async () => {});
     const page = { url: vi.fn(() => "http://127.0.0.1:9222/json/version") };

extensions/browser/src/browser/pw-tools-core.interactions.ts

@@ -12,6 +12,10 @@ import {
 } from "./act-policy.js";
 import type { BrowserActRequest, BrowserFormField } from "./client-actions.types.js";
 import { DEFAULT_FILL_FIELD_TYPE } from "./form-fields.js";
+import {
+  assertBrowserNavigationResultAllowed,
+  withBrowserNavigationPolicy,
+} from "./navigation-guard.js";
 import { DEFAULT_UPLOAD_DIR, resolveStrictExistingPathsWithinRoot } from "./paths.js";
 import {
   assertPageNavigationCompletedSafely,
@@ -119,20 +123,84 @@ function isMainFrameNavigation(page: NavigationObservablePage, frame: Frame): bo
   return frame === page.mainFrame();
 }
 
+async function assertSubframeNavigationAllowed(
+  frameUrl: string,
+  ssrfPolicy?: SsrFPolicy,
+): Promise<void> {
+  if (!ssrfPolicy || (!frameUrl.startsWith("http://") && !frameUrl.startsWith("https://"))) {
+    // Non-network frame URLs like about:blank and about:srcdoc do not cross the
+    // browser SSRF boundary, so they should not trigger the navigation policy.
+    return;
+  }
+
+  await assertBrowserNavigationResultAllowed({
+    url: frameUrl,
+    ...withBrowserNavigationPolicy(ssrfPolicy),
+  });
+}
+
+type ObservedDelayedNavigations = {
+  mainFrameNavigated: boolean;
+  subframes: string[];
+};
+
+function snapshotNetworkFrameUrl(frame: Frame): string | null {
+  try {
+    const frameUrl = frame.url();
+    return frameUrl.startsWith("http://") || frameUrl.startsWith("https://") ? frameUrl : null;
+  } catch {
+    return null;
+  }
+}
+
+async function assertObservedDelayedNavigations(opts: {
+  cdpUrl: string;
+  page: Page;
+  ssrfPolicy?: SsrFPolicy;
+  targetId?: string;
+  observed: ObservedDelayedNavigations;
+}): Promise<void> {
+  let subframeError: unknown;
+  try {
+    for (const frameUrl of opts.observed.subframes) {
+      await assertSubframeNavigationAllowed(frameUrl, opts.ssrfPolicy);
+    }
+  } catch (err) {
+    subframeError = err;
+  }
+  if (opts.observed.mainFrameNavigated) {
+    await assertPageNavigationCompletedSafely({
+      cdpUrl: opts.cdpUrl,
+      page: opts.page,
+      response: null,
+      ssrfPolicy: opts.ssrfPolicy,
+      targetId: opts.targetId,
+    });
+  }
+  if (subframeError) {
+    throw subframeError;
+  }
+}
+
 function observeDelayedInteractionNavigation(
   page: NavigationObservablePage,
   previousUrl: string,
-): Promise<boolean> {
+): Promise<ObservedDelayedNavigations> {
   if (didCrossDocumentUrlChange(page, previousUrl)) {
-    return Promise.resolve(true);
+    return Promise.resolve({ mainFrameNavigated: true, subframes: [] });
   }
   if (typeof page.on !== "function" || typeof page.off !== "function") {
-    return Promise.resolve(false);
+    return Promise.resolve({ mainFrameNavigated: false, subframes: [] });
   }
 
-  return new Promise<boolean>((resolve) => {
+  return new Promise<ObservedDelayedNavigations>((resolve) => {
+    const subframes: string[] = [];
     const onFrameNavigated = (frame: Frame) => {
       if (!isMainFrameNavigation(page, frame)) {
+        const frameUrl = snapshotNetworkFrameUrl(frame);
+        if (frameUrl) {
+          subframes.push(frameUrl);
+        }
         return;
       }
       // Use isHashOnlyNavigation rather than !didCrossDocumentUrlChange: the
@@ -142,11 +210,14 @@ function observeDelayedInteractionNavigation(
         return;
       }
       cleanup();
-      resolve(true);
+      resolve({ mainFrameNavigated: true, subframes });
     };
     const timeout = setTimeout(() => {
       cleanup();
-      resolve(didCrossDocumentUrlChange(page, previousUrl));
+      resolve({
+        mainFrameNavigated: didCrossDocumentUrlChange(page, previousUrl),
+        subframes,
+      });
     }, INTERACTION_NAVIGATION_GRACE_MS);
     const cleanup = () => {
       clearTimeout(timeout);
@@ -196,8 +267,13 @@ function scheduleDelayedInteractionNavigationGuard(opts: {
       }
       resolve();
     };
+    const subframes: string[] = [];
     const onFrameNavigated = (frame: Frame) => {
       if (!isMainFrameNavigation(page, frame)) {
+        const frameUrl = snapshotNetworkFrameUrl(frame);
+        if (frameUrl) {
+          subframes.push(frameUrl);
+        }
         return;
       }
       // Use isHashOnlyNavigation rather than !didCrossDocumentUrlChange: the
@@ -207,16 +283,26 @@ function scheduleDelayedInteractionNavigationGuard(opts: {
         return;
       }
       cleanup();
-      void assertPageNavigationCompletedSafely({
+      void assertObservedDelayedNavigations({
         cdpUrl: opts.cdpUrl,
         page: opts.page,
-        response: null,
         ssrfPolicy: opts.ssrfPolicy,
         targetId: opts.targetId,
+        observed: { mainFrameNavigated: true, subframes },
       }).then(() => settle(), settle);
     };
     const timeout = setTimeout(() => {
-      settle();
+      cleanup();
+      void assertObservedDelayedNavigations({
+        cdpUrl: opts.cdpUrl,
+        page: opts.page,
+        ssrfPolicy: opts.ssrfPolicy,
+        targetId: opts.targetId,
+        observed: {
+          mainFrameNavigated: didCrossDocumentUrlChange(page, opts.previousUrl),
+          subframes,
+        },
+      }).then(() => settle(), settle);
     }, INTERACTION_NAVIGATION_GRACE_MS);
     const cleanup = () => {
       clearTimeout(timeout);
@@ -248,8 +334,13 @@ async function assertInteractionNavigationCompletedSafely<T>(opts: {
   // slow interactions, silently bypassing the SSRF guard.
   const navPage = opts.page as unknown as NavigationObservablePage;
   let navigatedDuringAction = false;
+  const subframeNavigationsDuringAction: string[] = [];
   const onFrameNavigated = (frame: Frame) => {
     if (!isMainFrameNavigation(navPage, frame)) {
+      const frameUrl = snapshotNetworkFrameUrl(frame);
+      if (frameUrl) {
+        subframeNavigationsDuringAction.push(frameUrl);
+      }
       return;
     }
     // Use isHashOnlyNavigation rather than didCrossDocumentUrlChange: the event
@@ -278,6 +369,15 @@ async function assertInteractionNavigationCompletedSafely<T>(opts: {
   const navigationObserved =
     navigatedDuringAction || didCrossDocumentUrlChange(opts.page, opts.previousUrl);
 
+  let subframeError: unknown;
+  try {
+    for (const frameUrl of subframeNavigationsDuringAction) {
+      await assertSubframeNavigationAllowed(frameUrl, opts.ssrfPolicy);
+    }
+  } catch (err) {
+    subframeError = err;
+  }
+
   if (navigationObserved) {
     await assertPageNavigationCompletedSafely({
       cdpUrl: opts.cdpUrl,
@@ -290,17 +390,14 @@ async function assertInteractionNavigationCompletedSafely<T>(opts: {
     // Preserve the action-error path semantics: if a rejected click/evaluate still
     // triggers a delayed navigation, the SSRF block must win over the original
     // action error instead of surfacing a stale interaction failure.
-    const delayedNavigationObserved = await observeDelayedInteractionNavigation(
-      opts.page,
-      opts.previousUrl,
-    );
-    if (delayedNavigationObserved) {
-      await assertPageNavigationCompletedSafely({
+    const observed = await observeDelayedInteractionNavigation(opts.page, opts.previousUrl);
+    if (observed.mainFrameNavigated || observed.subframes.length > 0) {
+      await assertObservedDelayedNavigations({
         cdpUrl: opts.cdpUrl,
         page: opts.page,
-        response: null,
         ssrfPolicy: opts.ssrfPolicy,
         targetId: opts.targetId,
+        observed,
       });
     }
   } else {
@@ -316,6 +413,10 @@ async function assertInteractionNavigationCompletedSafely<T>(opts: {
     });
   }
 
+  if (subframeError) {
+    throw subframeError;
+  }
+
   if (actionError) {
     throw actionError;
   }
@@ -379,25 +480,6 @@ function createAbortPromiseWithListener(
     },
   };
 }
-
-async function assertPostInteractionNavigationSafe(opts: {
-  cdpUrl: string;
-  page: Awaited<ReturnType<typeof getPageForTargetId>>;
-  ssrfPolicy?: SsrFPolicy;
-  targetId?: string;
-}): Promise<void> {
-  if (!opts.ssrfPolicy) {
-    return;
-  }
-  await assertPageNavigationCompletedSafely({
-    cdpUrl: opts.cdpUrl,
-    page: opts.page,
-    response: null,
-    ssrfPolicy: opts.ssrfPolicy,
-    targetId: opts.targetId,
-  });
-}
-
 export async function highlightViaPlaywright(opts: {
   cdpUrl: string;
   targetId?: string;
@@ -559,12 +641,16 @@ export async function pressKeyViaPlaywright(opts: {
   }
   const page = await getPageForTargetId(opts);
   ensurePageState(page);
-  await page.keyboard.press(key, {
-    delay: Math.max(0, Math.floor(opts.delayMs ?? 0)),
-  });
-  await assertPostInteractionNavigationSafe({
+  const previousUrl = page.url();
+  await assertInteractionNavigationCompletedSafely({
+    action: async () => {
+      await page.keyboard.press(key, {
+        delay: Math.max(0, Math.floor(opts.delayMs ?? 0)),
+      });
+    },
     cdpUrl: opts.cdpUrl,
     page,
+    previousUrl,
     ssrfPolicy: opts.ssrfPolicy,
     targetId: opts.targetId,
   });
@@ -582,7 +668,7 @@ export async function typeViaPlaywright(opts: {
   ssrfPolicy?: SsrFPolicy;
 }): Promise<void> {
   const resolved = requireRefOrSelector(opts.ref, opts.selector);
-  const text = String(opts.text ?? "");
+  const text = opts.text ?? "";
   const page = await getRestoredPageForTarget(opts);
   const label = resolved.ref ?? resolved.selector!;
   const locator = resolved.ref
@@ -597,10 +683,14 @@ export async function typeViaPlaywright(opts: {
       await locator.fill(text, { timeout });
     }
     if (opts.submit) {
-      await locator.press("Enter", { timeout });
-      await assertPostInteractionNavigationSafe({
+      const previousUrl = page.url();
+      await assertInteractionNavigationCompletedSafely({
+        action: async () => {
+          await locator.press("Enter", { timeout });
+        },
         cdpUrl: opts.cdpUrl,
         page,
+        previousUrl,
         ssrfPolicy: opts.ssrfPolicy,
         targetId: opts.targetId,
       });

extensions/browser/src/browser/pw-tools-core.snapshot.navigate-guard.test.ts

@@ -7,6 +7,21 @@ import {
   setPwToolsCoreCurrentPage,
 } from "./pw-tools-core.test-harness.js";
 
+vi.mock("openclaw/plugin-sdk/browser-security-runtime", async () => {
+  const actual = await vi.importActual<
+    typeof import("openclaw/plugin-sdk/browser-security-runtime")
+  >("openclaw/plugin-sdk/browser-security-runtime");
+  const lookupFn = async (_hostname: string, options?: { all?: boolean }) => {
+    const result = { address: "93.184.216.34", family: 4 };
+    return options?.all === true ? [result] : result;
+  };
+  return {
+    ...actual,
+    resolvePinnedHostnameWithPolicy: (hostname: string, params: object = {}) =>
+      actual.resolvePinnedHostnameWithPolicy(hostname, { ...params, lookupFn: lookupFn as never }),
+  };
+});
+
 installPwToolsCoreTestHooks();
 const mod = await import("./pw-tools-core.snapshot.js");
 

extensions/browser/src/browser/pw-tools-core.snapshot.ts

@@ -89,7 +89,7 @@ export async function snapshotAiViaPlaywright(opts: {
     timeout: Math.max(500, Math.min(60_000, Math.floor(opts.timeoutMs ?? 5000))),
     track: "response",
   });
-  let snapshot = String(result?.full ?? "");
+  let snapshot = result?.full ?? "";
   const maxChars = opts.maxChars;
   const limit =
     typeof maxChars === "number" && Number.isFinite(maxChars) && maxChars > 0
@@ -152,7 +152,7 @@ export async function snapshotRoleViaPlaywright(opts: {
       timeout: 5000,
       track: "response",
     });
-    const built = buildRoleSnapshotFromAiSnapshot(String(result?.full ?? ""), opts.options);
+    const built = buildRoleSnapshotFromAiSnapshot(result?.full ?? "", opts.options);
     storeRoleRefsForTarget({
       page,
       cdpUrl: opts.cdpUrl,
@@ -178,7 +178,7 @@ export async function snapshotRoleViaPlaywright(opts: {
       : page.locator(":root");
 
   const ariaSnapshot = await locator.ariaSnapshot();
-  const built = buildRoleSnapshotFromAriaSnapshot(String(ariaSnapshot ?? ""), opts.options);
+  const built = buildRoleSnapshotFromAriaSnapshot(ariaSnapshot ?? "", opts.options);
   storeRoleRefsForTarget({
     page,
     cdpUrl: opts.cdpUrl,

extensions/browser/src/browser/pw-tools-core.state.ts

@@ -10,7 +10,7 @@ export async function setOfflineViaPlaywright(opts: {
 }): Promise<void> {
   const page = await getPageForTargetId(opts);
   ensurePageState(page);
-  await page.context().setOffline(Boolean(opts.offline));
+  await page.context().setOffline(opts.offline);
 }
 
 export async function setExtraHTTPHeadersViaPlaywright(opts: {
@@ -36,8 +36,8 @@ export async function setHttpCredentialsViaPlaywright(opts: {
     await page.context().setHTTPCredentials(null);
     return;
   }
-  const username = String(opts.username ?? "");
-  const password = String(opts.password ?? "");
+  const username = opts.username ?? "";
+  const password = opts.password ?? "";
   if (!username) {
     throw new Error("username is required (or set clear=true)");
   }

extensions/browser/src/browser/pw-tools-core.storage.ts

@@ -99,7 +99,7 @@ export async function storageSetViaPlaywright(opts: {
 }): Promise<void> {
   const page = await getPageForTargetId(opts);
   ensurePageState(page);
-  const key = String(opts.key ?? "");
+  const key = opts.key;
   if (!key) {
     throw new Error("key is required");
   }
@@ -108,7 +108,7 @@ export async function storageSetViaPlaywright(opts: {
       const store = kind === "session" ? window.sessionStorage : window.localStorage;
       store.setItem(k, value);
     },
-    { kind: opts.kind, key, value: String(opts.value ?? "") },
+    { kind: opts.kind, key, value: opts.value },
   );
 }
 

extensions/browser/src/browser/pw-tools-core.test-harness.ts

@@ -42,12 +42,28 @@ const sessionMocks = vi.hoisted(() => ({
   rememberRoleRefsForTarget: vi.fn(() => {}),
 }));
 
+const navigationGuardMocks = vi.hoisted(() => ({
+  assertBrowserNavigationResultAllowed: vi.fn(async () => {}),
+  withBrowserNavigationPolicy: vi.fn((ssrfPolicy?: unknown) => ({ ssrfPolicy })),
+}));
+
 vi.mock("./pw-session.js", () => sessionMocks);
+vi.mock("./navigation-guard.js", async (importOriginal) => {
+  const actual = await importOriginal<Record<string, unknown>>();
+  return {
+    ...actual,
+    ...navigationGuardMocks,
+  };
+});
 
 export function getPwToolsCoreSessionMocks() {
   return sessionMocks;
 }
 
+export function getPwToolsCoreNavigationGuardMocks() {
+  return navigationGuardMocks;
+}
+
 export function setPwToolsCoreCurrentPage(page: Record<string, unknown> | null) {
   currentPage = page;
 }
@@ -70,5 +86,8 @@ export function installPwToolsCoreTestHooks() {
     for (const fn of Object.values(sessionMocks)) {
       fn.mockClear();
     }
+    for (const fn of Object.values(navigationGuardMocks)) {
+      fn.mockClear();
+    }
   });
 }

extensions/browser/src/browser/pw-tools-core.waits-next-download-saves-it.test.ts

@@ -130,6 +130,9 @@ describe("pw-tools-core", () => {
 
     const res = await p;
     const outPath = (vi.mocked(saveAs).mock.calls as unknown as Array<[string]>)[0]?.[0];
+    if (typeof outPath !== "string") {
+      throw new Error("download save path was not captured");
+    }
     return { res, outPath };
   }
 
@@ -284,8 +287,8 @@ describe("pw-tools-core", () => {
       path.join(path.sep, "tmp", "openclaw-preferred", "downloads"),
     );
     const expectedDownloadsTail = `${path.join("tmp", "openclaw-preferred", "downloads")}${path.sep}`;
-    expect(path.dirname(String(outPath))).toBe(expectedRootedDownloadsDir);
-    expect(path.basename(String(outPath))).toMatch(/-file\.bin$/);
+    expect(path.dirname(outPath)).toBe(expectedRootedDownloadsDir);
+    expect(path.basename(outPath)).toMatch(/-file\.bin$/);
     expect(path.normalize(res.path)).toContain(path.normalize(expectedDownloadsTail));
     expect(tmpDirMocks.resolvePreferredOpenClawTmpDir).toHaveBeenCalled();
   });
@@ -297,10 +300,10 @@ describe("pw-tools-core", () => {
       suggestedFilename: "../../../../etc/passwd",
     });
     expect(typeof outPath).toBe("string");
-    expect(path.dirname(String(outPath))).toBe(
+    expect(path.dirname(outPath)).toBe(
       path.resolve(path.join(path.sep, "tmp", "openclaw-preferred", "downloads")),
     );
-    expect(path.basename(String(outPath))).toMatch(/-passwd$/);
+    expect(path.basename(outPath)).toMatch(/-passwd$/);
     expect(path.normalize(res.path)).toContain(
       path.normalize(`${path.join("tmp", "openclaw-preferred", "downloads")}${path.sep}`),
     );

extensions/browser/src/browser/routes/agent.act.existing-session-navigation-guard.test.ts

@@ -134,7 +134,7 @@ describe("existing-session interaction navigation guard", () => {
     expect(chromeMcpMocks.pressChromeMcpKey).toHaveBeenCalledWith(
       expect.objectContaining({ key: "Enter" }),
     );
-    expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).toHaveBeenCalledTimes(4);
+    expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).toHaveBeenCalledTimes(6);
     expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).toHaveBeenNthCalledWith(
       1,
       expect.objectContaining({ url: "https://example.com" }),
@@ -151,6 +151,14 @@ describe("existing-session interaction navigation guard", () => {
       4,
       expect.objectContaining({ url: "https://example.com" }),
     );
+    expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).toHaveBeenNthCalledWith(
+      5,
+      expect.objectContaining({ url: "https://example.com" }),
+    );
+    expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).toHaveBeenNthCalledWith(
+      6,
+      expect.objectContaining({ url: "https://example.com" }),
+    );
   });
 
   it("rechecks the page url after delayed navigation-triggering interactions", async () => {
@@ -198,10 +206,153 @@ describe("existing-session interaction navigation guard", () => {
       await pending;
     })();
 
-    await expect(completion).rejects.toThrow("Unable to verify post-interaction navigation");
+    await expect(completion).rejects.toThrow("Unable to verify stable post-interaction navigation");
     expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).not.toHaveBeenCalled();
   });
 
+  it("fails closed when a later post-action probe becomes unreadable", async () => {
+    chromeMcpMocks.evaluateChromeMcpScript
+      .mockResolvedValueOnce("result" as never) // action evaluate
+      .mockResolvedValueOnce("https://example.com" as never) // location probe 1
+      .mockResolvedValueOnce(undefined as never) // location probe 2 - unreadable
+      .mockResolvedValueOnce(undefined as never) // location probe 3 - unreadable
+      .mockResolvedValueOnce(undefined as never); // follow-up probe - still unreadable
+
+    const handler = getActPostHandler();
+    const response = createBrowserRouteResponse();
+    const pending =
+      handler?.(
+        { params: {}, query: {}, body: { kind: "evaluate", fn: "() => 1" } },
+        response.res,
+      ) ?? Promise.resolve();
+    void pending.catch(() => {});
+    const completion = (async () => {
+      await vi.runAllTimersAsync();
+      await pending;
+    })();
+
+    await expect(completion).rejects.toThrow("Unable to verify stable post-interaction navigation");
+    expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).toHaveBeenCalledOnce();
+    expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).toHaveBeenCalledWith(
+      expect.objectContaining({ url: "https://example.com" }),
+    );
+  });
+
+  it("confirms stability via follow-up probe when URL changes on the last loop iteration", async () => {
+    // Probe 1 (action evaluate result): returns the action value
+    // Location probe 1 (0ms): fails (context churn)
+    // Location probe 2 (250ms): reads safe URL A
+    // Location probe 3 (500ms): reads safe URL B (late navigation)
+    // Follow-up probe (500ms later): reads URL B again → stable, success
+    chromeMcpMocks.evaluateChromeMcpScript
+      .mockResolvedValueOnce("result" as never) // action evaluate result
+      .mockRejectedValueOnce(new Error("context churn") as never) // location probe 1 fails
+      .mockResolvedValueOnce("https://example.com" as never) // location probe 2: URL A
+      .mockResolvedValueOnce("https://safe-redirect.com" as never) // location probe 3: URL B (changed)
+      .mockResolvedValueOnce("https://safe-redirect.com" as never); // follow-up: URL B again → stable
+
+    const response = await runAction({ kind: "evaluate", fn: "() => 1" });
+
+    expect(response.statusCode).toBe(200);
+    // 1 action call + 5 location probes (3 in loop + 1 failed + 1 follow-up)
+    expect(chromeMcpMocks.evaluateChromeMcpScript).toHaveBeenCalledTimes(5);
+    expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).toHaveBeenCalledTimes(3);
+    expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).toHaveBeenNthCalledWith(
+      1,
+      expect.objectContaining({ url: "https://example.com" }),
+    );
+    expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).toHaveBeenNthCalledWith(
+      2,
+      expect.objectContaining({ url: "https://safe-redirect.com" }),
+    );
+    expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).toHaveBeenNthCalledWith(
+      3,
+      expect.objectContaining({ url: "https://safe-redirect.com" }),
+    );
+  });
+
+  it("keeps probing through the full window before declaring navigation stable", async () => {
+    chromeMcpMocks.evaluateChromeMcpScript
+      .mockResolvedValueOnce("result" as never) // action evaluate result
+      .mockResolvedValueOnce("https://example.com" as never) // location probe 1
+      .mockResolvedValueOnce("https://example.com" as never) // location probe 2
+      .mockResolvedValueOnce("https://safe-redirect.com" as never) // location probe 3
+      .mockResolvedValueOnce("https://safe-redirect.com" as never); // follow-up confirms late redirect
+
+    const response = await runAction({ kind: "evaluate", fn: "() => 1" });
+
+    expect(response.statusCode).toBe(200);
+    expect(chromeMcpMocks.evaluateChromeMcpScript).toHaveBeenCalledTimes(5);
+    expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).toHaveBeenCalledTimes(4);
+    expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).toHaveBeenNthCalledWith(
+      1,
+      expect.objectContaining({ url: "https://example.com" }),
+    );
+    expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).toHaveBeenNthCalledWith(
+      2,
+      expect.objectContaining({ url: "https://example.com" }),
+    );
+    expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).toHaveBeenNthCalledWith(
+      3,
+      expect.objectContaining({ url: "https://safe-redirect.com" }),
+    );
+    expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).toHaveBeenNthCalledWith(
+      4,
+      expect.objectContaining({ url: "https://safe-redirect.com" }),
+    );
+  });
+
+  it("fails closed when follow-up probe sees yet another URL change", async () => {
+    chromeMcpMocks.evaluateChromeMcpScript
+      .mockResolvedValueOnce("result" as never) // action evaluate result
+      .mockResolvedValueOnce("https://a.com" as never) // location probe 1
+      .mockResolvedValueOnce("https://b.com" as never) // location probe 2: changed
+      .mockResolvedValueOnce("https://c.com" as never) // location probe 3: changed again
+      .mockResolvedValueOnce("https://d.com" as never); // follow-up: still changing
+
+    const handler = getActPostHandler();
+    const response = createBrowserRouteResponse();
+    const pending =
+      handler?.(
+        { params: {}, query: {}, body: { kind: "evaluate", fn: "() => 1" } },
+        response.res,
+      ) ?? Promise.resolve();
+    void pending.catch(() => {});
+    const completion = (async () => {
+      await vi.runAllTimersAsync();
+      await pending;
+    })();
+
+    await expect(completion).rejects.toThrow("Unable to verify stable post-interaction navigation");
+  });
+
+  it("fails closed when a probe error follows two stable reads", async () => {
+    // Probes 1 + 2 match (sawStableAllowedUrl would be true), probe 3 throws.
+    // Guard must NOT return success — the throw invalidates prior stability.
+    chromeMcpMocks.evaluateChromeMcpScript
+      .mockResolvedValueOnce("result" as never) // action evaluate result
+      .mockResolvedValueOnce("https://example.com" as never) // location probe 1
+      .mockResolvedValueOnce("https://example.com" as never) // location probe 2 → stable pair
+      .mockRejectedValueOnce(new Error("context destroyed") as never) // location probe 3 → error
+      .mockRejectedValueOnce(new Error("context destroyed") as never); // follow-up → still errored
+
+    const handler = getActPostHandler();
+    const response = createBrowserRouteResponse();
+    const pending =
+      handler?.(
+        { params: {}, query: {}, body: { kind: "evaluate", fn: "() => 1" } },
+        response.res,
+      ) ?? Promise.resolve();
+    void pending.catch(() => {});
+    const completion = (async () => {
+      await vi.runAllTimersAsync();
+      await pending;
+    })();
+
+    await expect(completion).rejects.toThrow("Unable to verify stable post-interaction navigation");
+    expect(navigationGuardMocks.assertBrowserNavigationResultAllowed).toHaveBeenCalledTimes(2);
+  });
+
   it("skips the guard when no SSRF policy is configured", async () => {
     const response = await runAction({ kind: "press", key: "Enter" }, null);
 

extensions/browser/src/browser/routes/agent.act.normalize.ts

@@ -318,4 +318,5 @@ export function normalizeActRequest(
       };
     }
   }
+  throw new Error("Unsupported browser act kind");
 }

extensions/browser/src/browser/routes/agent.act.ts

@@ -70,7 +70,6 @@ async function assertExistingSessionPostInteractionNavigationAllowed(params: {
   profileName: string;
   userDataDir?: string;
   targetId: string;
-  baselineUrl?: string;
   ssrfPolicy?: BrowserNavigationPolicyOptions["ssrfPolicy"];
 }): Promise<void> {
   const ssrfPolicyOpts = withBrowserNavigationPolicy(params.ssrfPolicy);
@@ -78,9 +77,8 @@ async function assertExistingSessionPostInteractionNavigationAllowed(params: {
     return;
   }
 
-  let lastUrl = params.baselineUrl ?? "";
-  let sawStableUrl = false;
-  let readSucceeded = false;
+  let lastObservedUrl: string | undefined;
+  let sawStableAllowedUrl = false;
   for (const delayMs of EXISTING_SESSION_INTERACTION_NAVIGATION_RECHECK_DELAYS_MS) {
     if (delayMs > 0) {
       await sleep(delayMs);
@@ -89,27 +87,49 @@ async function assertExistingSessionPostInteractionNavigationAllowed(params: {
     try {
       currentUrl = await readExistingSessionLocationHref(params);
     } catch {
+      sawStableAllowedUrl = false;
       continue;
     }
-    readSucceeded = true;
     await assertBrowserNavigationResultAllowed({
       url: currentUrl,
       ...ssrfPolicyOpts,
     });
-    if (currentUrl !== lastUrl) {
-      lastUrl = currentUrl;
-      sawStableUrl = false;
-      continue;
+    if (currentUrl === lastObservedUrl) {
+      sawStableAllowedUrl = true;
+    } else {
+      sawStableAllowedUrl = false;
     }
-    if (sawStableUrl) {
-      return;
-    }
-    sawStableUrl = true;
+    lastObservedUrl = currentUrl;
   }
 
-  if (!readSucceeded) {
-    throw new Error("Unable to verify post-interaction navigation");
+  if (sawStableAllowedUrl) {
+    return;
   }
+
+  // If the loop exhausted without confirming stability but we did observe
+  // at least one allowed URL, run a single follow-up probe so a late URL
+  // transition that has already settled is not treated as a false failure.
+  if (lastObservedUrl) {
+    const lastDelay =
+      EXISTING_SESSION_INTERACTION_NAVIGATION_RECHECK_DELAYS_MS[
+        EXISTING_SESSION_INTERACTION_NAVIGATION_RECHECK_DELAYS_MS.length - 1
+      ];
+    await sleep(lastDelay);
+    try {
+      const followUpUrl = await readExistingSessionLocationHref(params);
+      await assertBrowserNavigationResultAllowed({
+        url: followUpUrl,
+        ...ssrfPolicyOpts,
+      });
+      if (followUpUrl === lastObservedUrl) {
+        return;
+      }
+    } catch {
+      // Probe failed — fall through to throw
+    }
+  }
+
+  throw new Error("Unable to verify stable post-interaction navigation");
 }
 
 async function runExistingSessionActionWithNavigationGuard<T>(params: {
@@ -287,6 +307,7 @@ function getExistingSessionUnsupportedMessage(action: BrowserActRequest): string
     case "close":
       return null;
   }
+  throw new Error("Unsupported browser act kind");
 }
 
 export function registerBrowserAgentActRoutes(
@@ -351,7 +372,6 @@ export function registerBrowserAgentActRoutes(
             profileName,
             userDataDir: profileCtx.profile.userDataDir,
             targetId: tab.targetId,
-            baselineUrl: tab.url,
             ssrfPolicy,
           };
           const unsupportedMessage = getExistingSessionUnsupportedMessage(action);

extensions/browser/src/browser/routes/tabs.test.ts

@@ -0,0 +1,112 @@
+import { describe, expect, it, vi } from "vitest";
+import { registerBrowserTabRoutes } from "./tabs.js";
+import { createBrowserRouteApp, createBrowserRouteResponse } from "./test-helpers.js";
+
+function createProfileContext(overrides?: Partial<ReturnType<typeof baseProfileContext>>) {
+  return {
+    ...baseProfileContext(),
+    ...overrides,
+  };
+}
+
+function baseProfileContext() {
+  return {
+    profile: {
+      name: "openclaw",
+    },
+    ensureBrowserAvailable: vi.fn(async () => {}),
+    ensureTabAvailable: vi.fn(async () => ({
+      targetId: "T1",
+      title: "Tab 1",
+      url: "https://example.com",
+      type: "page",
+    })),
+    isHttpReachable: vi.fn(async () => true),
+    isReachable: vi.fn(async () => true),
+    listTabs: vi.fn(async () => [
+      {
+        targetId: "T1",
+        title: "Tab 1",
+        url: "https://example.com",
+        type: "page",
+      },
+    ]),
+    openTab: vi.fn(async () => ({
+      targetId: "T1",
+      title: "Tab 1",
+      url: "https://example.com",
+      type: "page",
+    })),
+    focusTab: vi.fn(async () => {}),
+    closeTab: vi.fn(async () => {}),
+    stopRunningBrowser: vi.fn(async () => ({ stopped: false })),
+    resetProfile: vi.fn(async () => ({ moved: false, from: "" })),
+  };
+}
+
+function createRouteContext(profileCtx: ReturnType<typeof createProfileContext>) {
+  return {
+    state: () => ({ resolved: { ssrfPolicy: undefined } }),
+    forProfile: () => profileCtx,
+    listProfiles: vi.fn(async () => []),
+    mapTabError: vi.fn(() => null),
+    ensureBrowserAvailable: profileCtx.ensureBrowserAvailable,
+    ensureTabAvailable: profileCtx.ensureTabAvailable,
+    isHttpReachable: profileCtx.isHttpReachable,
+    isReachable: profileCtx.isReachable,
+    listTabs: profileCtx.listTabs,
+    openTab: profileCtx.openTab,
+    focusTab: profileCtx.focusTab,
+    closeTab: profileCtx.closeTab,
+    stopRunningBrowser: profileCtx.stopRunningBrowser,
+    resetProfile: profileCtx.resetProfile,
+  };
+}
+
+async function callTabsAction(params: {
+  body: Record<string, unknown>;
+  profileCtx: ReturnType<typeof createProfileContext>;
+}) {
+  const { app, postHandlers } = createBrowserRouteApp();
+  registerBrowserTabRoutes(app, createRouteContext(params.profileCtx) as never);
+  const handler = postHandlers.get("/tabs/action");
+  expect(handler).toBeTypeOf("function");
+
+  const response = createBrowserRouteResponse();
+  await handler?.({ params: {}, query: {}, body: params.body }, response.res);
+  return response;
+}
+
+describe("browser tab routes", () => {
+  it("returns browser-not-running for close when the browser is not reachable", async () => {
+    const profileCtx = createProfileContext({
+      isReachable: vi.fn(async () => false),
+    });
+
+    const response = await callTabsAction({
+      body: { action: "close", index: 0 },
+      profileCtx,
+    });
+
+    expect(response.statusCode).toBe(409);
+    expect(response.body).toEqual({ error: "browser not running" });
+    expect(profileCtx.listTabs).not.toHaveBeenCalled();
+    expect(profileCtx.closeTab).not.toHaveBeenCalled();
+  });
+
+  it("returns browser-not-running for select when the browser is not reachable", async () => {
+    const profileCtx = createProfileContext({
+      isReachable: vi.fn(async () => false),
+    });
+
+    const response = await callTabsAction({
+      body: { action: "select", index: 0 },
+      profileCtx,
+    });
+
+    expect(response.statusCode).toBe(409);
+    expect(response.body).toEqual({ error: "browser not running" });
+    expect(profileCtx.listTabs).not.toHaveBeenCalled();
+    expect(profileCtx.focusTab).not.toHaveBeenCalled();
+  });
+});

extensions/browser/src/browser/routes/tabs.ts

@@ -201,6 +201,9 @@ export function registerBrowserTabRoutes(app: BrowserRouteRegistrar, ctx: Browse
         }
 
         if (action === "close") {
+          if (!(await ensureBrowserRunning(profileCtx, res))) {
+            return;
+          }
           const tabs = await profileCtx.listTabs();
           const target = resolveIndexedTab(tabs, index);
           if (!target) {
@@ -214,6 +217,9 @@ export function registerBrowserTabRoutes(app: BrowserRouteRegistrar, ctx: Browse
           if (typeof index !== "number") {
             return jsonError(res, 400, "index is required");
           }
+          if (!(await ensureBrowserRunning(profileCtx, res))) {
+            return;
+          }
           const tabs = await profileCtx.listTabs();
           const target = tabs[index];
           if (!target) {

extensions/browser/src/browser/screenshot.test.ts

@@ -22,8 +22,8 @@ describe("browser screenshot normalization", () => {
 
     expect(normalized.buffer.byteLength).toBeLessThanOrEqual(5 * 1024 * 1024);
     const meta = await sharp(normalized.buffer).metadata();
-    expect(Number(meta.width)).toBeLessThanOrEqual(2000);
-    expect(Number(meta.height)).toBeLessThanOrEqual(2000);
+    expect(meta.width).toBeLessThanOrEqual(2000);
+    expect(meta.height).toBeLessThanOrEqual(2000);
     expect(normalized.buffer[0]).toBe(0xff);
     expect(normalized.buffer[1]).toBe(0xd8);
   }, 120_000);

extensions/browser/src/browser/screenshot.ts

@@ -19,8 +19,8 @@ export async function normalizeBrowserScreenshot(
   const maxBytes = Math.max(1, Math.round(opts?.maxBytes ?? DEFAULT_BROWSER_SCREENSHOT_MAX_BYTES));
 
   const meta = await getImageMetadata(buffer);
-  const width = Number(meta?.width ?? 0);
-  const height = Number(meta?.height ?? 0);
+  const width = meta?.width ?? 0;
+  const height = meta?.height ?? 0;
   const maxDim = Math.max(width, height);
 
   if (buffer.byteLength <= maxBytes && (maxDim === 0 || (width <= maxSide && height <= maxSide))) {

extensions/browser/src/browser/server-context.existing-session.test.ts

@@ -1,7 +1,22 @@
 import fs from "node:fs";
-import { beforeEach, describe, expect, it, vi } from "vitest";
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
 import type { BrowserServerState } from "./server-context.js";
 
+vi.mock("openclaw/plugin-sdk/browser-security-runtime", async () => {
+  const actual = await vi.importActual<
+    typeof import("openclaw/plugin-sdk/browser-security-runtime")
+  >("openclaw/plugin-sdk/browser-security-runtime");
+  const lookupFn = async (_hostname: string, options?: { all?: boolean }) => {
+    const result = { address: "93.184.216.34", family: 4 };
+    return options?.all === true ? [result] : result;
+  };
+  return {
+    ...actual,
+    resolvePinnedHostnameWithPolicy: (hostname: string, params: object = {}) =>
+      actual.resolvePinnedHostnameWithPolicy(hostname, { ...params, lookupFn: lookupFn as never }),
+  };
+});
+
 vi.mock("./chrome-mcp.js", () => ({
   closeChromeMcpSession: vi.fn(async () => true),
   ensureChromeMcpAvailable: vi.fn(async () => {}),
@@ -59,9 +74,23 @@ function makeState(): BrowserServerState {
 }
 
 beforeEach(() => {
+  for (const key of [
+    "ALL_PROXY",
+    "all_proxy",
+    "HTTP_PROXY",
+    "http_proxy",
+    "HTTPS_PROXY",
+    "https_proxy",
+  ]) {
+    vi.stubEnv(key, "");
+  }
   vi.clearAllMocks();
 });
 
+afterEach(() => {
+  vi.unstubAllEnvs();
+});
+
 describe("browser server-context existing-session profile", () => {
   it("routes tab operations through the Chrome MCP backend", async () => {
     fs.mkdirSync("/tmp/brave-profile", { recursive: true });

extensions/browser/src/browser/server-context.loopback-direct-ws.test.ts

@@ -1,6 +1,7 @@
 import { afterEach, describe, expect, it, vi } from "vitest";
 import { withFetchPreconnect } from "../../test-support.js";
 import * as cdpModule from "./cdp.js";
+import { BrowserCdpEndpointBlockedError } from "./errors.js";
 import { createBrowserRouteContext } from "./server-context.js";
 import { makeState, originalFetch } from "./server-context.remote-tab-ops.harness.js";
 
@@ -139,4 +140,25 @@ describe("browser server-context loopback direct WebSocket profiles", () => {
     await openclaw.focusTab("T2");
     await openclaw.closeTab("T2");
   });
+
+  it("blocks direct WebSocket tab operations when strict SSRF policy rejects the cdpUrl", async () => {
+    const fetchMock = vi.fn(async () => {
+      throw new Error("unexpected fetch");
+    });
+
+    global.fetch = withFetchPreconnect(fetchMock);
+    const state = makeState("openclaw");
+    state.resolved.ssrfPolicy = { dangerouslyAllowPrivateNetwork: false };
+    state.resolved.profiles.openclaw = {
+      cdpUrl: "ws://10.0.0.42:18800/devtools/browser/SESSION?token=abc",
+      color: "#FF4500",
+    };
+    const ctx = createBrowserRouteContext({ getState: () => state });
+    const openclaw = ctx.forProfile("openclaw");
+
+    await expect(openclaw.listTabs()).rejects.toBeInstanceOf(BrowserCdpEndpointBlockedError);
+    await expect(openclaw.focusTab("T1")).rejects.toBeInstanceOf(BrowserCdpEndpointBlockedError);
+    await expect(openclaw.closeTab("T1")).rejects.toBeInstanceOf(BrowserCdpEndpointBlockedError);
+    expect(fetchMock).not.toHaveBeenCalled();
+  });
 });

extensions/browser/src/browser/server-context.remote-profile-tab-ops.playwright.test.ts

@@ -36,15 +36,16 @@ describe("browser remote profile tab ops via Playwright", () => {
     expect(opened.targetId).toBe("T2");
     expect(state.profiles.get("remote")?.lastTargetId).toBe("T2");
     expect(createPageViaPlaywright).toHaveBeenCalledWith({
-      cdpUrl: "https://browserless.example/chrome?token=abc",
+      cdpUrl: "https://1.1.1.1:9222/chrome?token=abc",
       url: "http://127.0.0.1:3000",
       ssrfPolicy: { allowPrivateNetwork: true },
     });
 
     await remote.closeTab("T1");
     expect(closePageByTargetIdViaPlaywright).toHaveBeenCalledWith({
-      cdpUrl: "https://browserless.example/chrome?token=abc",
+      cdpUrl: "https://1.1.1.1:9222/chrome?token=abc",
       targetId: "T1",
+      ssrfPolicy: { allowPrivateNetwork: true },
     });
     expect(fetchMock).not.toHaveBeenCalled();
   });
@@ -140,13 +141,54 @@ describe("browser remote profile tab ops via Playwright", () => {
 
     await remote.focusTab("T1");
     expect(focusPageByTargetIdViaPlaywright).toHaveBeenCalledWith({
-      cdpUrl: "https://browserless.example/chrome?token=abc",
+      cdpUrl: "https://1.1.1.1:9222/chrome?token=abc",
       targetId: "T1",
+      ssrfPolicy: { allowPrivateNetwork: true },
     });
     expect(fetchMock).not.toHaveBeenCalled();
     expect(state.profiles.get("remote")?.lastTargetId).toBe("T1");
   });
 
+  it("blocks remote Playwright tab operations when strict SSRF policy rejects the cdpUrl", async () => {
+    const listPagesViaPlaywright = vi.fn(async () => [
+      { targetId: "T1", title: "Tab 1", url: "https://example.com", type: "page" },
+    ]);
+    const focusPageByTargetIdViaPlaywright = vi.fn(async () => {});
+    const closePageByTargetIdViaPlaywright = vi.fn(async () => {});
+
+    vi.spyOn(deps.pwAiModule, "getPwAiModule").mockResolvedValue({
+      listPagesViaPlaywright,
+      focusPageByTargetIdViaPlaywright,
+      closePageByTargetIdViaPlaywright,
+    } as unknown as Awaited<ReturnType<typeof deps.pwAiModule.getPwAiModule>>);
+
+    const state = deps.makeState("remote");
+    state.resolved.ssrfPolicy = { dangerouslyAllowPrivateNetwork: false };
+    state.resolved.profiles.remote = {
+      ...state.resolved.profiles.remote,
+      cdpUrl: "http://10.0.0.42:9222",
+      cdpPort: 9222,
+    };
+    const ctx = deps.createBrowserRouteContext({ getState: () => state });
+    const remote = ctx.forProfile("remote");
+
+    await expect(remote.listTabs()).rejects.toMatchObject({
+      name: "BrowserCdpEndpointBlockedError",
+      status: 400,
+    });
+    await expect(remote.focusTab("T1")).rejects.toMatchObject({
+      name: "BrowserCdpEndpointBlockedError",
+      status: 400,
+    });
+    await expect(remote.closeTab("T1")).rejects.toMatchObject({
+      name: "BrowserCdpEndpointBlockedError",
+      status: 400,
+    });
+    expect(listPagesViaPlaywright).not.toHaveBeenCalled();
+    expect(focusPageByTargetIdViaPlaywright).not.toHaveBeenCalled();
+    expect(closePageByTargetIdViaPlaywright).not.toHaveBeenCalled();
+  });
+
   it("does not swallow Playwright runtime errors for remote profiles", async () => {
     vi.spyOn(deps.pwAiModule, "getPwAiModule").mockResolvedValue({
       listPagesViaPlaywright: vi.fn(async () => {

extensions/browser/src/browser/server-context.remote-tab-ops.harness.ts

@@ -17,7 +17,7 @@ export function makeState(
       cdpPortRangeStart: 18800,
       cdpPortRangeEnd: 18899,
       cdpProtocol: profile === "remote" ? "https" : "http",
-      cdpHost: profile === "remote" ? "browserless.example" : "127.0.0.1",
+      cdpHost: profile === "remote" ? "1.1.1.1" : "127.0.0.1",
       cdpIsLoopback: profile !== "remote",
       remoteCdpTimeoutMs: 1500,
       remoteCdpHandshakeTimeoutMs: 3000,
@@ -31,8 +31,8 @@ export function makeState(
       defaultProfile: profile,
       profiles: {
         remote: {
-          cdpUrl: "https://browserless.example/chrome?token=abc",
-          cdpPort: 443,
+          cdpUrl: "https://1.1.1.1:9222/chrome?token=abc",
+          cdpPort: 9222,
           color: "#00AA00",
         },
         openclaw: { cdpPort: 18800, color: "#FF4500" },