fix(agent): use provider registry for claw402, echo reasoning_content for thinking models, add Beta badge

- Agent now uses mcp.NewAIClientByProvider() for claw402 provider, ensuring
  x402 payment signing works correctly instead of generic HTTP client
- Added ReasoningContent field to Message/LLMResponse structs and wired
  serialization/parsing so DeepSeek thinking models work in multi-turn
- Added Beta badge to Agent nav tab in HeaderBar

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

agent/agent.go

@@ -165,13 +165,26 @@ func (a *Agent) loadAIClientFromStoreUser(storeUserID string) (mcp.AIClient, str
 			apiKey := strings.TrimSpace(string(model.APIKey))
 			customAPIURL := strings.TrimSpace(model.CustomAPIURL)
 			modelName := strings.TrimSpace(model.CustomModelName)
-			customAPIURL, modelName = resolveModelRuntimeConfig(model.Provider, customAPIURL, modelName, model.ID)
+			provider := strings.ToLower(strings.TrimSpace(model.Provider))
+
+			// Use the provider registry for providers like claw402 that have their own
+			// client implementation (x402 payment, custom auth, etc.).
+			if client := mcp.NewAIClientByProvider(provider); client != nil {
+				if modelName == "" {
+					modelName = model.ID
+				}
+				client.SetAPIKey(apiKey, customAPIURL, modelName)
+				a.log().Info("agent AI client selected (provider registry)", "store_user_id", candidateUserID, "model_id", model.ID, "provider", provider, "model", modelName)
+				return client, modelName, true
+			}
+
+			customAPIURL, modelName = resolveModelRuntimeConfig(provider, customAPIURL, modelName, model.ID)
 			if apiKey == "" || customAPIURL == "" {
 				a.log().Warn(
 					"skipping incomplete enabled AI model",
 					"store_user_id", candidateUserID,
 					"model_id", model.ID,
-					"provider", model.Provider,
+					"provider", provider,
 					"has_api_key", apiKey != "",
 					"has_custom_api_url", customAPIURL != "",
 				)

agent/planner_runtime.go

@@ -4014,6 +4014,9 @@ func (a *Agent) thinkAndActLegacyWithStore(ctx context.Context, storeUserID stri
 		if resp.Content != "" {
 			assistantMsg.Content = resp.Content
 		}
+		if resp.ReasoningContent != "" {
+			assistantMsg.ReasoningContent = resp.ReasoningContent
+		}
 		messages = append(messages, assistantMsg)
 
 		for _, tc := range resp.ToolCalls {

mcp/client.go

@@ -278,8 +278,9 @@ func (client *Client) ParseMCPResponseFull(body []byte) (*LLMResponse, error) {
 	var result struct {
 		Choices []struct {
 			Message struct {
-				Content   string     `json:"content"`
+				Content          string     `json:"content"`
-				ToolCalls []ToolCall `json:"tool_calls"`
+				ReasoningContent string     `json:"reasoning_content"`
+				ToolCalls        []ToolCall `json:"tool_calls"`
 			} `json:"message"`
 		} `json:"choices"`
 		Usage struct {
@@ -310,8 +311,9 @@ func (client *Client) ParseMCPResponseFull(body []byte) (*LLMResponse, error) {
 
 	msg := result.Choices[0].Message
 	return &LLMResponse{
-		Content:   msg.Content,
+		Content:          msg.Content,
-		ToolCalls: msg.ToolCalls,
+		ReasoningContent: msg.ReasoningContent,
+		ToolCalls:        msg.ToolCalls,
 	}, nil
 }
 
@@ -624,6 +626,11 @@ func (client *Client) BuildRequestBodyFromRequest(req *Request) map[string]any {
 		} else {
 			m["content"] = msg.Content
 		}
+		// DeepSeek thinking models require reasoning_content to be echoed back
+		// in multi-turn conversations when present in assistant messages.
+		if msg.ReasoningContent != "" {
+			m["reasoning_content"] = msg.ReasoningContent
+		}
 		messages = append(messages, m)
 	}
 

mcp/request.go

@@ -6,10 +6,11 @@ import "context"
 // Supports plain messages (Role+Content), assistant tool-call messages (ToolCalls),
 // and tool result messages (Role="tool", ToolCallID, Content).
 type Message struct {
-	Role       string     `json:"role"`                  // "system", "user", "assistant", "tool"
+	Role             string     `json:"role"`                            // "system", "user", "assistant", "tool"
-	Content    string     `json:"content,omitempty"`     // Text content (omitted when ToolCalls present)
+	Content          string     `json:"content,omitempty"`               // Text content (omitted when ToolCalls present)
-	ToolCalls  []ToolCall `json:"tool_calls,omitempty"`  // Set by assistant when calling tools
+	ReasoningContent string     `json:"reasoning_content,omitempty"`     // Thinking-model reasoning (must be echoed back in multi-turn)
-	ToolCallID string     `json:"tool_call_id,omitempty"` // Set on role="tool" result messages
+	ToolCalls        []ToolCall `json:"tool_calls,omitempty"`            // Set by assistant when calling tools
+	ToolCallID       string     `json:"tool_call_id,omitempty"`          // Set on role="tool" result messages
 }
 
 // ToolCall is a single function call requested by the LLM.
@@ -29,8 +30,9 @@ type ToolCallFunction struct {
 // text reply (Content) and any structured tool calls (ToolCalls).
 // Exactly one of the two fields will be non-empty for a well-formed response.
 type LLMResponse struct {
-	Content   string     // Plain-text reply (final answer)
+	Content          string     // Plain-text reply (final answer)
-	ToolCalls []ToolCall // Structured tool invocations
+	ReasoningContent string     // Thinking-model reasoning content
+	ToolCalls        []ToolCall // Structured tool invocations
 }
 
 // Tool represents a tool/function that AI can call

web/src/components/common/HeaderBar.tsx

@@ -115,6 +115,7 @@ export default function HeaderBar({
                   page: 'agent',
                   path: ROUTES.agent,
                   label: 'Agent',
+                  badge: 'Beta',
                   requiresAuth: false,
                 },
                 {