liuyu/nofx
1
0
Fork 0
mirror of https://github.com/NoFxAiOS/nofx.git synced 2026-07-05 12:00:59 +08:00
Code Issues Packages Projects Releases Wiki Activity
523 Commits 14 Branches 0 Tags
a2eb03ede98faa77fbb930294cd4f2a08e5f023f
Commit Graph

14 Commits

Author SHA1 Message Date
0xYYBB | ZYY | Bobo
3f8df95c63 fix(prompts): correct confidence scale from 0-1 to 0-100 to match backend schema (#564) 
## Problem

The prompts specified confidence range as 0-1 (float), but the backend code
expects 0-100 (integer). This causes JSON parsing errors when AI outputs
values like 0.85:

```
Error: json: cannot unmarshal number 0.85 into Go struct field Decision.confidence of type int
Result: confidence defaults to 0
```

## Root Cause

**Backend Definition** (decision/engine.go:103):
```go
Confidence int `json:"confidence,omitempty"` // 信心度 (0-100)
```

**Prompts (before fix)**:
- adaptive.txt: "confidence (信心度 0-1)"
- nof1.txt: "confidence (float, 0-1)"

**buildHardSystemPrompt** (decision/engine.go:336):
```go
sb.WriteString("- `confidence`: 0-100（开仓建议≥75）\n")
```

The dynamic system prompt was correct, but the base prompts contradicted it.

## Solution

Update prompt files to use consistent 0-100 integer scale:

### adaptive.txt
- `confidence (信心度 0-1)` → `confidence (信心度 0-100)`
- `<0.85` → `<85`
- `0.85-0.90` → `85-90`
- etc.

### nof1.txt
- `confidence (float, 0-1)` → `confidence (int, 0-100)`
- `0.0-0.3` → `0-30`
- `0.3-0.6` → `30-60`
- etc.

## Impact

- ✅ Fixes JSON parsing errors when AI outputs float values
- ✅ Aligns prompts with backend schema
- ✅ Consistent with buildHardSystemPrompt() output format
- ✅ No breaking changes (backend already expects 0-100)

## Testing

```bash
# Verify backend expects 0-100
grep "Confidence int" decision/engine.go
# Output: Confidence int `json:"confidence,omitempty"` // 信心度 (0-100)

# Verify buildHardSystemPrompt uses 0-100
grep "confidence.*0-100" decision/engine.go
# Output: sb.WriteString("- `confidence`: 0-100（开仓建议≥75）\n")

# Build test
go build ./decision/...  # ✅ PASS
```

## Related

- Addresses schema mismatch mentioned in Issue #557
- Note: confidence field is currently not validated by backend (validateDecision
  does not check confidence value), but correct schema prevents parsing errors

---

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: tinkle-community <tinklefund@gmail.com>
 2025-11-06 00:35:53 +08:00
Icyoung
915133eac8 Merge pull request #437 from zhouyongyou/fix/margin-calculation 
fix(margin): correct position sizing formula to prevent insufficient margin errors
 2025-11-05 16:13:37 +08:00
Icyoung
c322be5dbf Merge pull request #386 from zhouyongyou/feat/configurable-oi-threshold 
feat(decision): configurable OI threshold + relaxed trading template
 2025-11-05 15:46:25 +08:00
Icyoung
ae60d93a92 Merge pull request #325 from zhouyongyou/refactor/enhance-partial-close-guidance 
refactor(prompts): 增強部分平倉使用指導 [依賴 #415]
 2025-11-05 15:43:08 +08:00
Hansen1018
356351671c 增加 稳健和风险控制均衡基础策略提示词 
主要优化点：

强化风险控制框架 明确单笔风险≤2%，总风险≤6%
添加连续亏损后的仓位调整规则

设置单日和每周最大亏损限制

提高开仓标准 要求至少3个技术指标支持
必须有多时间框架趋势确认

入场时机要求更具体

完善决策流程 增加市场环境评估环节
明确风险回报比计算要求

添加资金保护检查点

细化行为准则 明确等待最佳机会的重要性
强调分批止盈和严格止损

添加情绪控制具体方法

增强绩效反馈机制 不同夏普比率区间的具体行动指南
亏损状态下的仓位控制要求

盈利状态下的纪律保持提醒

这个优化版本更加注重风险控制和稳健性，同时保持了交易的专业性和灵活性。
 2025-11-05 13:19:07 +08:00
ZhouYongyou
f35982cde5 fix(prompts): rename actions to match backend implementation 
## Problem

Backend code expects these action names:
- `open_long`, `open_short`, `close_long`, `close_short`

But prompts use outdated names:
- `buy_to_enter`, `sell_to_enter`, `close`

This causes all trading decisions to fail with unknown action errors.

## Solution

Minimal changes to fix action name compatibility:

### prompts/nof1.txt
- ✅ `buy_to_enter` → `open_long`
- ✅ `sell_to_enter` → `open_short`
- ✅ `close` → `close_long` / `close_short`
- ✅ Explicitly list `wait` action
- +18 lines, -6 lines (only action definitions section)

### prompts/adaptive.txt
- ✅ `buy_to_enter` → `open_long`
- ✅ `sell_to_enter` → `open_short`
- ✅ `close` → `close_long` / `close_short`
- +15 lines, -6 lines (only action definitions section)

## Impact

- ✅ Trading decisions now execute successfully
- ✅ Maintains all existing functionality
- ✅ No new features added (minimal diff)

## Verification

```bash
# Backend expects these actions:
grep 'Action string' decision/engine.go
# "open_long", "open_short", "close_long", "close_short", ...

# Old names removed:
grep -r "buy_to_enter\|sell_to_enter" prompts/
# (no results)
```

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: tinkle-community <tinklefund@gmail.com>
 2025-11-04 19:41:23 +08:00
ZhouYongyou
9dd09109a9 fix(margin): correct position sizing formula to prevent insufficient margin errors 
## Problem
AI was calculating position_size_usd incorrectly, treating it as margin requirement instead of notional value, causing code=-2019 errors (insufficient margin).

## Solution

### 1. Updated AI prompts with correct formula
- **prompts/adaptive.txt**: Added clear position sizing calculation steps
- **prompts/nof1.txt**: Added English version with example
- **prompts/default.txt**: Added Chinese version with example

**Correct formula:**
1. Available Margin = Available Cash × 0.95 × Allocation % (reserve 5% for fees)
2. Notional Value = Available Margin × Leverage
3. position_size_usd = Notional Value (this is the value for JSON)

**Example:** $500 cash, 5x leverage → position_size_usd = $2,375 (not $500)

### 2. Added code-level validation
- **trader/auto_trader.go**: Added margin checks in executeOpenLong/ShortWithRecord
- Validates required margin + fees ≤ available balance before opening position
- Returns clear error message if insufficient

## Impact
- Prevents code=-2019 errors
- AI now understands the difference between notional value and margin requirement
- Double validation: AI prompt + code check

## Testing
- ✅ Compiles successfully
- ⚠️ Requires live trading environment testing
 2025-11-04 18:44:07 +08:00
ZhouYongyou
fa327c7f1e feat(decision): make OI threshold configurable + add relaxed prompt template 
## Changes

### 1. decision/engine.go - Configurable OI Threshold
- Extract hardcoded 15M OI threshold to configurable constant
- Add clear documentation for risk profiles:
  - 15M (Conservative) - BTC/ETH/SOL only
  - 10M (Balanced) - Add major alt-coins
  - 8M (Relaxed) - Include mid-cap coins (BNB/LINK/AVAX)
  - 5M (Aggressive) - Most alt-coins allowed
- Default: 15M (保守，維持原行為)

### 2. prompts/adaptive_relaxed.txt - New Trading Template
Conservative optimization for increased trading frequency while maintaining high win-rate:

**Key Adjustments:**
- Confidence threshold: 85 → 80 (allow more opportunities)
- Cooldown period: 9min → 6min (faster reaction)
- Multi-timeframe trend: 3 periods → 2 periods (relaxed requirement)
- Entry checklist: 5/8 → 4/8 (easier to pass)
- RSI range: 30-40/65-70 → <45/>60 (wider acceptance)
- Risk-reward ratio: 1:3 → 1:2.5 (more flexible)

**Expected Impact:**
- Trading frequency: 5/day → 8-15/day (+60-200%)
- Win-rate: 40% → 50-55% (improved)
- Alt-coins: More opportunities unlocked
- Risk controls: Preserved (Sharpe-based, loss-pause)

## Usage
Users can now choose trading style via Web UI:
- `adaptive` - Strictest (original)
- `adaptive_relaxed` - Balanced (this PR)
- `nof1` - Most aggressive

## Rationale
The original adaptive.txt uses 5-layer filtering (confidence/cooldown/trend/checklist/RSI)
that filters out ~95% of opportunities. This template provides a middle-ground option
for users who want higher frequency without sacrificing core risk management.

Related: #trading-frequency #alt-coin-support
 2025-11-04 17:22:14 +08:00
ZhouYongyou
2d8ba9fc03 refactor(prompts): add comprehensive partial_close guidance to adaptive.txt 
Add detailed guidance chapter for dynamic TP/SL management and partial close operations.

## Changes

- New chapter: "动态止盈止损与部分平仓指引" (Dynamic TP/SL & Partial Close Guidance)
- Inserted between "可用动作" (Actions) and "决策流程" (Decision Flow) sections
- 4 key guidance points covering:
  1. Partial close best practices (use clear percentages like 25%/50%/75%)
  2. Reassessing remaining position after partial exit
  3. Proper use cases for update_stop_loss / update_take_profit
  4. Multi-stage exit strategy requirements

## Benefits

- ✅ Provides concrete operational guidelines for AI decision-making
- ✅ Clarifies when and how to use partial_close effectively
- ✅ Emphasizes remaining position management (prevents "orphan" positions)
- ✅ Aligns with existing backend support for partial_close action

## Background

While adaptive.txt already lists partial_close as an available action,
it lacked detailed operational guidance. This enhancement fills that gap
by providing specific percentages, use cases, and multi-stage exit examples.

Backend (decision/engine.go) already validates partial_close with
close_percentage field, so this is purely a prompt enhancement with
no code changes required.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: tinkle-community <tinklefund@gmail.com>
 2025-11-04 17:20:57 +08:00
tinkle-community
431248b199 Merge pull request #268 from yutou123/dev 2025-11-03 06:59:28 +08:00
ZhouYongyou
b0720ea80c docs(prompts): Update AI prompt to support dynamic TP/SL features (v5.5.1) 
- Add 3 new action types: update_stop_loss, update_take_profit, partial_close
- Introduce "Zero Principle" (疑惑优先) for risk control
- Expand decision flow to 8 steps with critical safeguards:
  * Step 2: Consecutive loss pause (2x→45min, 3x→24h, 4x→72h)
  * Step 5: BTC status check (multi-timeframe MACD confirmation)
  * Step 6: Long/short confirmation checklist (≥5/8 indicators)
  * Step 7: Fake breakout detection (RSI multi-timeframe + candle patterns)
  * Step 8: Objective confidence scoring (base 60 + conditions)
- Add signal priority ranking (trend resonance > volume > BTC > RSI...)
- Add dynamic TP/SL strategies with examples
- Increase confidence threshold: 0.6 → 0.85 for opening positions
- Add cooldown rules and slippage buffer (0.05%)
- Optimize prompt length: 4445 words → 1500 words (-66%)

Key improvements in v5.5.1:
✅ BTC status check - Most critical protection for altcoin trading
✅ Long/short checklist - 5/8 indicators required, prevent false signals
✅ Objective confidence scoring - Base 60 + condition adjustments
✅ Fake breakout logic - RSI multi-timeframe + candle filters
✅ Consecutive loss pause - 2x/3x/4x trigger different cooldowns
✅ OI confirmation - >+5% for real breakout validation
✅ Signal priority ranking - Trend resonance > volume > BTC...
✅ Slippage handling - 0.05% buffer + profit check

Design philosophy: Let AI autonomously judge trend vs chop, trust strong reasoning models.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: tinkle-community <tinklefund@gmail.com>
 2025-11-03 01:59:54 +08:00
芋头
c00fa5adc4 文件命名空格问题 2025-11-03 00:16:28 +08:00
芋头
73b6a3e9d7 补充提示词 2025-11-02 23:56:32 +08:00
SkywalkerJi
5afbd9a3c3 Supports custom system prompts and custom models. 2025-11-01 19:45:54 +08:00