--- name: okx-compound-strategy description: > AI-driven compound trading strategy v3.6.0 with Price Action + MACD convergence. Phase 0.5: market context (Supertrend, ADX, BBWidth, HV, MFI, funding, L/S ratio, sentiment). Phase 0.6: Gemma4:e4b macro analysis (Daily/12H/4H/1H trends, S&R levels, DIRECTION). BTC-USDT-SWAP: 50x isolated, bidirectional (LONG/SHORT/FLAT per Gemma4). Entry: 8-dimension 100-pt scoring + Pinbar auto-detection + hot-spot confluence bonus. Exit: dynamic TP thresholds (BULLISH/NEUTRAL/BEARISH) + MACD 30m divergence TP + overextension mean-reversion TP + support-break SL. RAVE-USDT-SWAP: short with 6-signal detection incl. MACD head-and-shoulders top. Grid: dual-scan (gainers + OI), 7-dimension scoring, funding-rate avoidance. Self-iteration authorized within defined limits. All orders tagged agentTradeKit. license: MIT metadata: author: compound-strategy version: "3.6.0" category: compound-strategy dependencies: okx-cex-market, okx-cex-trade, okx-cex-bot, okx-cex-portfolio --- # OKX Compound Strategy This skill depends on and orchestrates the following OKX official skills: - **okx-cex-market** — market data, technical indicators, OI - **okx-cex-trade** — perpetual swap order placement - **okx-cex-bot** — contract grid bot management - **okx-cex-portfolio** — account balance and positions Install dependencies first: ```bash npx skills add okx/agent-skills ``` **CRITICAL**: All order placement commands MUST include `--tag agentTradeKit` for leaderboard counting. --- ## Capital Allocation | Strategy | Budget | Instrument | Leverage | Mode | |----------|--------|------------|----------|------| | BTC Trend | 400 USDT margin | BTC-USDT-SWAP | 50x isolated | Long-biased | | RAVE Dedicated Short | Up to 200 USDT (uses BTC budget if BTC inactive) | RAVE-USDT-SWAP | 10x isolated | Short only | | Volatility Grid | 100 USDT total | Top-5 gainer SWAPs | 10x isolated | Long or Short grid | --- ## When to Use This Skill Execute this skill every 5 minutes. Each cycle: 1. Load persistent state from `state.json` 2. Run account check 3. **Run market context assessment** — volatility + sentiment + capital flow 4. Run BTC strategy phase (AI confidence-scored entry) 5. Run RAVE dedicated short phase (optional, signal-gated) 6. Run volatility grid strategy phase (dual OI+price scan) 7. Save updated state to `state.json` All okx CLI commands must be run with: ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx ... ``` State file schema: ```json { "peak_equity": 0, "btc_activated": false, "gemma4_analysis": { "trend_daily": "", "trend_12h": "", "trend_4h": "", "trend_1h": "", "support_1": 0, "support_2": 0, "resistance_1": 0, "resistance_2": 0, "direction": "", "confidence": "", "updated_at": "" }, "btc_position": { "active": false, "entry_count": 0, "avg_entry_price": 0, "total_sz": 0, "remaining_pct": 100 }, "rave_position": { "active": false, "avg_entry_price": 0, "total_sz": 0, "remaining_pct": 100, "signal_count": 0, "swing_high": 0, "vegas_addon_count": 0, "bb_addon_count": 0 }, "grid_bots": [ { "algoId": "", "instId": "", "direction": "", "maxPx": 0, "minPx": 0, "sl_px": 0, "lever": 10, "sz": 0 } ], "paused_grids": [ { "instId": "", "direction": "", "maxPx": 0, "minPx": 0, "sl_px": 0, "lever": 10, "sz": 0, "resume_after": "" } ] } ``` --- ## PHASE 0.1 — AI综合决策层 (v3.14, 比赛合规核心) **此策略由Claude (AI) 作为核心决策者运行**, 规则评分只是AI的**输入参考**, **最终决定永远由AI做出**, 可override任何机械规则。 ### AI决策原则 每个开仓/平仓决策, AI必须: 1. 收集所有输入信号 (评分/结构/context) 2. 综合考虑 (不是简单加权平均) 3. 输出binding decision + rationale 4. 允许override机械规则 (记录为什么) ### Binding Decision格式 ``` [AI DECISION {instId} {timestamp}] Input signals: - MTF Confluence score: {n}/100 - Macro context: {state} - Structure: {description} - Volatility regime: {regime} Mechanical rule says: {GO/SKIP/HOLD} AI synthesis: {3-5 sentences reasoning, integrating quantitative + qualitative} Final decision: {action} Override mechanical?: {yes/no and why} Position params: size: {margin} leverage: {x} SL: {price, based on structure} TP: {price, based on structure} Risk: {worst case loss} Confidence: {0-1} ``` ### AI override scenarios - 评分65分但结构破位 → override to SKIP - 评分40分但多时间框架一致 → override to GO (small size) - TP规则说平仓但momentum仍强 → override to HOLD - SL规则触发但price action示反转 → override to HOLD 1 more bar --- ## PHASE 0.2 — Macro Bear Filter (v3.14, grid protection) **来源**: 今日EDGE/THETA grid在单边下跌市死扛导致-$20损失 ### 规则 每cycle必须check: ```python btc_1h_supertrend = get_supertrend("BTC", "1H") btc_30m_last3 = last_3_30m_bars # close < open count macro_bearish = ( btc_1h_supertrend == "DOWN" AND sum(b.close < b.open for b in btc_30m_last3) >= 2 ) ``` ### Grid响应 (若macro_bearish): | 当前grid状态 | 行动 | |---|---| | 新grid创建请求 | **拒绝** (等macro转好) | | 已运行grid pnl > 0% | 保持 + trail 30% drawdown | | 已运行grid pnl -5%~0% | 警戒 + tighten trail | | 已运行grid pnl < -5% | **立即stop** (不等-10%) | ### 解除条件 `BTC 1H Supertrend UP持续2小时` AND `30m连续2根阳线` → 恢复正常grid逻辑 --- ## PHASE 0.3 — Leverage Pre-Check (v3.14, 强制) **来源**: BTC杠杆错误3次 (10x vs 设计50x) ### 每次place新仓前必须run: ```bash actual_lever=$(okx swap get-leverage --instId ${INST_ID} --mgnMode isolated | grep short | awk '{print $4}') if [ "$actual_lever" != "$EXPECTED_LEVER" ]; then okx swap leverage --instId ${INST_ID} --lever ${EXPECTED_LEVER} --mgnMode isolated --posSide short # re-verify verify_lever fi log "LEVERAGE CHECK: ${INST_ID} lever=${EXPECTED_LEVER} ✓" ``` ### 各标的默认 - BTC = **50x** (永远) - ETH = **50x** - RAVE = **5-10x** (高波动, 不用50x) - Altcoins动态扫描 = **5x** - Grid = **10x** --- ## PHASE 0.4 — 动态单边趋势标的扫描（v3.8, 每 cycle 必跑） **改造原因 (v3.7 → v3.8)**: v3.7 仅用 24h 涨跌幅 + 24h 位置筛选，但"24h跌很多"不等于"现在还在单边下跌"。 很多币24h跌20%但最近2小时已经筑底反弹。这种"尸体"挂空单=送钱。 v3.8 增加**多指标趋势确认**，只选择真正处于单边趋势中的标的。 ### 第一步：初筛 (基于24h数据 + 成交额) ``` 粗筛门槛: vol > 50M USDT (流动性保障) AND (24h跌幅>12% 或 24h涨幅>10%) (放宽阈值, 因有后续趋势filter) ``` 输出初筛候选池（预计每cycle 5-20个标的）。 ### 第二步：趋势确认栈 (对每个初筛候选, 并行拉取指标) 需要拉取的指标： - **1H 均线**: price / EMA20 / EMA50 (对齐判断) - **1H Supertrend**: direction (UP/DOWN) - **1H ADX + DI+/DI-**: 趋势强度与方向动能 - **5m 最近5根K线 close**: 短线动量一致性 ### 第三步：质量分级 (对每个候选) **SHORT 质量等级**: | 等级 | 条件 (需全部满足) | 执行动作 | |------|------------------|----------| | **A级 满仓** | 1) 1H price < EMA20 < EMA50 (均线空头排列)
2) 1H Supertrend = DOWN
3) 1H ADX > 25 (强趋势)
4) 1H DI- > DI+ (空方动能强)
5) 5m最近5根 ≥4根阴线 | **市价SHORT** (40U margin 5x)
OCO: TP -15% / SL +8% | | **B级 限价** | 满足A级中4项 | 限价SHORT @ 当前×1.03
(30U margin 5x)
OCO: TP -12% / SL +10% | | **C级 轻仓** | 满足A级中3项 (必含均线排列+Supertrend) | 限价SHORT @ 当前×1.05
(20U margin 5x)
OCO: TP -10% / SL +12% | | 跳过 | 不满足C级 | 不入场 | **LONG 质量等级** (镜像逻辑): | 等级 | 条件 (需全部满足) | 执行动作 | |------|------------------|----------| | **A级 满仓** | 1) 1H price > EMA20 > EMA50
2) 1H Supertrend = UP
3) 1H ADX > 25
4) 1H DI+ > DI-
5) 5m最近5根 ≥4根阳线 | **市价LONG** (40U margin 5x)
OCO: TP +15% / SL -8% | | **B级 限价** | 满足A级中4项 | 限价LONG @ 当前×0.97
(30U margin 5x)
OCO: TP +12% / SL -10% | | **C级 轻仓** | 满足A级中3项 (必含均线排列+Supertrend) | 限价LONG @ 当前×0.95
(20U margin 5x)
OCO: TP +10% / SL -12% | | 跳过 | 不满足C级 | 不入场 | ### 第四步：执行与去重 **执行顺序**: 1. 按质量等级排序 (A > B > C) 2. 最多同时持有 **2个A级市价单 + 2个B级限价 + 2个C级限价** 3. 同标的已有持仓/挂单 → 跳过 4. 总风险上限: 所有新仓+挂单potential SL loss ≤ 当前equity的 8% **持仓后监控** (每cycle重新评估): - 若持仓标的 1H Supertrend 翻向 → 立即市价平仓 (趋势反转, 不等SL) - 若持仓标的 1H ADX 跌破 20 → 降级为trail stop (趋势转震荡) - 若 5m 出现强反转K线 (bull-engulf/three-soldiers on short 或反向) → 减仓50% ### 关键哲学 - **均线 = 真实趋势的骨架**, 比24h涨跌幅有效10倍 - **ADX > 25 = 有效趋势**, 否则只是噪音 - **多指标共振 = 高胜率**, 单一指标 = 赌博 - **A级市价直接打**, 不等待反弹 (真单边趋势不会给好价位) - **C级限价远离**, 给足缓冲等回踩 - **持仓后trend翻转立即平**, 纪律优先 --- ## PHASE 0.4.1 — ★Multi-TF Confluence 交叉验证★ (v3.13, 用户核心教学) **用户精髓**: "多级别K线结合起来看, 交叉验证, 胜率更高" ### 5级时间框架分工 | 周期 | 任务 | 核心判断 | |------|------|----------| | Daily / 6h | 大周期骨架 | Vegas通道排列, 长期趋势结构 | | 4h / 1h | 趋势验证 | MACD方向, RSI趋势, ADX强度 | | 30m / 15m | 结构确认 | Swing高低序列, Lower Highs/Higher Lows | | 5m | 动能确认 | BB位置, 反弹力度, volume | | 1m | 精准入场 | 反弹衰竭, pinbar, volume climax | ### Confluence 评分规则 (每级20分) **Daily/6h (+20)**: ``` +20: 方向一致 (空仓+空排列 或 多仓+多排列) +10: 中性 (均线粘合) 0: 方向矛盾 (空仓+多排列) ``` **4h/1h (+20)**: ``` +20: MACD方向 + RSI趋势 + Supertrend 全一致 +15: 三者中2个一致 +10: 仅1个一致 0: 全矛盾 ``` **30m/15m (+20)**: ``` +20: 清晰Lower Highs (空) 或 Higher Lows (多) 序列 +15: 有序列但有1次违反 +10: 序列不清晰 0: 反向序列 ``` **5m (+20)**: ``` +20: BB/RSI/MACD动能与方向一致, 不在极端反转区 +15: 2/3一致 +10: 仅1个一致 0: 全矛盾 ``` **1m (+20)**: ``` +20: 清晰入场信号 (pinbar反转 / 量能衰竭 / 突破回踩) +15: 有信号但不强 +10: 横盘无信号 0: 反向信号 ``` ### 入场决策 (按总分) | 分数 | 等级 | 操作 | 仓位 | |------|------|------|------| | ≥ 85 | A+ | 满仓市价进 (不等反弹) | 50U margin | | 70-84 | A | 半仓市价进 + 限价加仓 | 30U | | 55-69 | B | 限价单轻仓 (等更好时机) | 15U | | 40-54 | C | 观察 (不入场) | 0 | | < 40 | SKIP | 不入场 | 0 | ### 持仓决策 (已有仓) ``` 分数 ≥ 60 = 结构仍支持 → 继续持有 分数 40-59 = 结构转弱 → 考虑trailing tight / 减仓50% 分数 < 40 = 结构反转 → 立即平仓出场 ``` ### 每cycle必跑 (新加到流程最前) ``` cycle开始 → 1. Phase 1: 账户对账 2. Phase 0.4.1: Multi-TF Confluence评分 (BTC + 持仓品种) 3. Phase 0.4.2-0.4.6: 其他结构判断 4. Phase 0.5: 市场context 5. Phase 0.6: Gemma4 ... 必输出: [MTF CONFLUENCE {instId}] Daily/6h: +{n}/20 — {one line reasoning} 4h/1h: +{n}/20 — {reasoning} 30m/15m: +{n}/20 — {reasoning} 5m: +{n}/20 — {reasoning} 1m: +{n}/20 — {reasoning} ───────── Total: {n}/100 = {A+/A/B/C/SKIP} Direction: {LONG/SHORT/NONE} Suggested action: {description} ``` --- ## PHASE 0.4.2 — Vegas Channel三均线系统 (v3.12, 2026-04-19 用户6h图分析) **来源**: 用户6h BTC分析用 Vegas Wave 144/169/233 作为"结构生死线"。 ### 三均线配置 ``` EMA144 (橙) — 中期趋势线 EMA169 (蓝) — 中期确认线 EMA233 (红) — ★长期生死线★ ``` ### 结构判断规则 ``` 排列状态: 多头排列: EMA144 > EMA169 > EMA233 + 价格 > EMA233 空头排列: EMA144 < EMA169 < EMA233 + 价格 < EMA233 粘合/混乱: 三线间距 < 2% → 等方向 交易方向过滤: - 空头排列 → 只做空 - 多头排列 → 只做多 - 粘合 → 观望 (不操作) 单边确认: - 价格突破EMA233 = 长期方向改变 - 三线角度加速发散 = 趋势强度加强 ``` ### 实现 (每cycle计算) ```bash # 对BTC-USDT-SWAP在6H bar级别计算 python3 -c " closes = [读取最近250根6h K线 close] k144 = 2/(144+1); k169 = 2/(169+1); k233 = 2/(233+1) ema144 = closes[0] ema169 = closes[0] ema233 = closes[0] for c in closes[1:]: ema144 = c*k144 + ema144*(1-k144) ema169 = c*k169 + ema169*(1-k169) ema233 = c*k233 + ema233*(1-k233) print(f'price={closes[-1]} ema144={ema144:.0f} ema169={ema169:.0f} ema233={ema233:.0f}') # 输出判断: bull/bear/neutral " ``` --- ## PHASE 0.4.3 — "小实体连续阴线"识别 (v3.12, 用户盘感量化) **来源**: 用户6h分析"小实体连续阴线 = 下跌中继, 后续大跌概率大"。 ### Continuation Contraction Pattern **公式**: ```python # 取最近 N+3 根K线 (N=10, 给背景均值) bars = last_13_candles # 0=newest # Body = |close - open| body = lambda b: abs(b.close - b.open) # 最近3根实体 b0, b1, b2 = body(bars[0]), body(bars[1]), body(bars[2]) # 背景均实体 (前10根的平均) avg_body = mean([body(b) for b in bars[3:13]]) is_contraction = ( # 条件1: 连续3根阴线 all(bars[i].close < bars[i].open for i in [0,1,2]) # 条件2: 实体递减 AND b0 < b1 < b2 # 条件3: 实体显著小于背景 AND b0 < avg_body * 0.6 ) ``` ### 识别后的行动 ``` 识别到 Continuation Contraction: → 信号权重: 空头+8分 (进入评分系统) → 预期目标: current × (1 - 最近大阴线幅度 × 1.5) → 建议动作: 市价SHORT小仓 + 挂限价在轻微反弹位加仓 若识别后K线形态被破坏 (出现大阳线): → 立即取消所有pending短单 → 考虑结构反转 ``` ### 适用时间框架 - 强信号: 4h / 6h / 12h / Daily (大周期结构) - 弱信号: 15m / 30m / 1H (小周期, 假信号多) --- ## PHASE 0.4.6 — 目标价测算 + 时间止损 (v3.12, 用户教学) ### 目标价公式 (幅度对称性) ```python # 找最近10根K线中最大阴线的相对跌幅 last_10_bars = last_10_on_6h # 或其他合适周期 bearish_bars = [b for b in last_10_bars if b.close < b.open] max_drop_pct = max([(b.open - b.close)/b.open for b in bearish_bars]) # 目标价 = 当前 × (1 - max_drop × 1.5) projected_target = current_price × (1 - max_drop_pct × 1.5) ``` ### 时间止损 (time-based exit) ``` 每个结构性做空/做多持仓设 time_target: entry_time + N * bar_period (N=2-3 bars) 例: BTC 6h短, 预期2-3根6h K线到目标 entry 15:12 + 12h = 次日3:12 到时间未达预期: - 若仍有浮盈 → 考虑保本出场 - 若浮亏 → 立即市价平 - 若在预期区间但未到TP → 继续持有 逻辑: 结构性判断有"保质期", 超时未验证 = 分析失效, 止损保本 ``` --- ## PHASE 0.4.3 — 结构化TP/SL原则（v3.11, 2026-04-19） **用户教学**: "止赢止损的设置，除了根据张跌幅设置，还可以加入考虑前高、前低的位置，综合考虑。" ### 核心原则 不再用简单百分比（如SL +2%, TP -5%）设止盈止损, 而是**锚定结构位**： | 元素 | 锚点选择 | 逻辑 | |------|---------|------| | **SL (空单)** | 最近swing high + 0.3% | 价格破前高 = 下跌结构失效 | | **SL (多单)** | 最近swing low - 0.3% | 价格破前低 = 上升结构失效 | | **TP1 (空单)** | 最近intraday low | 首个目标, 常被测试 | | **TP1 (多单)** | 最近intraday high | 首个目标 | | **TP2 (空单)** | 24h low / 前日low | 中期目标, 更深测试 | | **TP2 (多单)** | 24h high / 前日high | 中期目标 | | **TP3 (激进)** | BB lowerBand / 关键斐波0.786 | 极限目标 | ### 例子 (RAVE 2026-04-19 15:25) 数据: - H1=1.534 (recent peak) / H2=1.444 / H3=1.248 (最新swing高) - 今日low=1.10 (capitulation) - 24h low=0.941 - 5m BB lower=1.204, 15m BB lower=0.933 限价SHORT @ 1.35 (fib 0.618 bounce): - ~~百分比法~~: SL +10%=1.485 / TP -25%=1.013 - **结构化法**: SL = H2 1.444 × 1.01 = **1.46** (前高0.01上方) TP = 今日low **1.10** (测试回补) - R/R: (1.35-1.10)/(1.46-1.35) = 0.25/0.11 = **2.27** ### 组合优势 1. **止损更合理**: 基于"如果结构破坏我错了"而非"我能亏多少%" 2. **止盈更精准**: 前低/前高是真实支撑阻力, 常被测试 3. **可视化交易**: 画线能看到SL/TP在图上的含义 ### 决策顺序 ``` 1. 确定方向 (多/空) 2. 找最近swing high和swing low (5m或15m级别) 3. 设SL = 结构破坏位 + 0.3%缓冲 4. 设TP1 = 反向最近swing点 5. 验证R/R ≥ 1.8 (否则skip) 6. 用户教的"激进"情景下, 可加TP2/TP3分层出场 ``` ### 失效条件 若swing high/low太远 (距离当前价>5%), 且百分比SL会更紧, 则: - 优先用结构位SL (宁可止损点远) - 缩小仓位来控制亏损总额 - 不用百分比SL强行拉近 (容易被洗) --- ## PHASE 0.4.4 — 下跌结构中的"Lower High 衰减"规则（v3.10, 用户盘感量化） **教训来源**: 用户指出挂空@75700"显然不合理"——因为75776已是前高，下一个反弹peak应更低。 ### 核心原则 在确认的下跌趋势中： - 每次反弹的peak **必然低于**前次反弹的peak - 即 H_N < H_{N-1} < H_{N-2} (单调递减) - 若反弹突破前高 → **下跌结构被打破**，策略失效 ### 量化公式 **Step 1 — 识别最近3个Swing High**: 在30m或15m K线上找最近3个显著高点 (HH pattern): - H1 = 最远的high（约1-2小时前） - H2 = 中间的high - H3 = 最近的high (刚形成) **Step 2 — 计算下一反弹peak预期 H4**: ``` 保守版: H4 = H3 × 0.996 (再低0.4%) 衰减版: H4 = H3 - (H2-H3) × 0.7 (按历史衰减幅度外推) ``` 取两者中较低的为entry参考。 **Step 3 — 入场与止损**: ``` 入场区: H4 × 0.995 ~ H4 × 1.005 (±0.5%缓冲) 止损SL: H3 × 1.003 (前高上方0.3%, 不是远远的H1) 止盈TP: 接近近期low × 0.98 (向下测试新低) ``` ### 验证反弹有效性（避免"反弹失败"的尴尬） ``` 若当前价格 < H4 × 0.97 (反弹力度不足到H4区) → 价格直接走空头续跌, 可考虑: - 破位市价短 (如果破前低确认) - OR 等下一次反弹机会 ``` ### 失效触发 ``` 若价格突破 H3 (前高) → 下降结构被破坏, 取消所有空单挂单 → 重新评估方向 ``` ### 多时间框架组合 ``` 30m → 确认下跌结构 (H1>H2>H3) 5m → 判断当前处于哪一段 (反弹/下跌/盘整) 1m → 精准入场 (微观反弹peak见顶) ``` ### 例子 (BTC 2026-04-19): - H1 ≈ 76500, H2 ≈ 76100, H3 ≈ 75776 (30m swing highs, 单调递减✓) - 衰减幅度: 76500→76100 (-400), 76100→75776 (-324) - 下一衰减预期: -324×0.7 = -227 - **H4 预计 ≈ 75776 - 227 = 75549** - 入场区: 75500-75600 - SL: 75776 × 1.003 = 76003 (round 76000) - TP: 74000-74400 (前期低点区) --- ## PHASE 0.4.5 — 高波动标的专属规则（v3.9, 2026-04-19实盘教训驱动） **教训来源**: RAVE 1.10空单被1.25 SL洗出(-$21), 13.6%SL对53%日内波动币种=送钱。 ### 规则1 — 高波动识别 ``` 高波动标的定义: 24h波动(high-low)/avg > 30% (而非24h涨跌) 或 24h change绝对值 > 40% ``` ### 规则2 — 高波动专用SL距离 ``` 标的类型 SL距离 备注 普通(<20%波动) 1-2×ATR 或 3-5% 标准 中波动(20-30%) 2-3×ATR 或 8-12% 适中 高波动(>30%) 3-5×ATR 或 **20-30%** RAVE类 极端(>50%) 5×ATR 或 **30-50%** 必须宽 ``` **例子**: RAVE 24h波动>50%, 入场1.10的SL应在 **1.55 (+40%)**, 不是1.25(+13.6%). ### 规则3 — 阶梯挂单 (Layered Orders) 对高波动单边标的, 同时挂**3个限价单**: ``` SHORT例子 (价格现价$1.10): 单1: limit SHORT @ 1.15 sz=10张 (靠近市价, 捕捉小反弹) 单2: limit SHORT @ 1.30 sz=10张 (中反弹) 单3: limit SHORT @ 1.50 sz=10张 (大反弹/诱多顶) 共同OCO: TP: 当前价 × 0.8 (大目标) SL: 当前价 × 1.6 (远stop, 给足空间) 优势: - 任一fill都有后续支援 - 反弹越大, 成交越多, 平均价越高 (空单更优) - SL统一设在1.6倍价位 (真趋势反转才触发) ``` ### 规则4 — Trailing SL for 高波动持仓 ``` 持仓后按利润梯度移SL: profit > 5% → SL移到 breakeven (入场价) profit > 10% → SL移到 入场价 × 0.97 (锁3%) profit > 20% → SL移到 入场价 × 0.90 (锁10%) profit > 40% → trailing: peak × 1.1 (peak上浮10%) 每次 SL调整通过 okx swap algo amend 执行, 保持OCO有效. ``` ### 规则5 — Rolling Orders (SL触发后自动复位) ``` 条件: 限价单被SL止损后, 以下全部仍成立: 1. 标的24h波动仍 > 30% 2. Phase 0.4趋势指标仍对齐单边方向 3. 距离上次入场距离 > 25% (避免反复洗) 行动: 立即在更远的价位挂新的阶梯限价单 例: RAVE@1.10入场, SL@1.55触发, RAVE pump到1.55+ → 立即挂 @ 1.70 / 1.90 / 2.10 三层新单 → 给SHORT继续进场机会 ``` ### 规则6 — 高波动标的仓位上限 ``` 单个高波动标的总margin ≤ 60U (vs 普通的30U) - 因为SL范围大, 固定亏损金额不变时仓位应缩小 - 例: SL距离30% × 5x = 150%浮亏 → margin=20U, 最大损失=$30 ``` ### 综合决策逻辑（AI binding） ``` 当Phase 0.4扫到高波动标的 (24h波动>30%): 1. 用规则2确定SL距离 (不用默认×1.15) 2. 用规则3挂阶梯 (不挂单层) 3. 持仓后用规则4做trailing 4. SL后用规则5评估是否re-entry 5. 全程遵守规则6仓位上限 ``` --- ## PHASE 0.5 — Market Context Assessment (every cycle, before any trade) Gather the following in parallel: ```bash # Trend direction & strength NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator supertrend BTC-USDT-SWAP --bar 1H --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator adx BTC-USDT-SWAP --bar 1H --profile tradebot # Volatility regime NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator bbwidth BTC-USDT-SWAP --bar 1H --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator hv BTC-USDT-SWAP --bar 1H --profile tradebot # Capital flow NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator mfi BTC-USDT-SWAP --bar 1H --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market funding-rate BTC-USDT-SWAP --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator top-long-short BTC-USDT-SWAP --profile tradebot # Sentiment NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx news coin-sentiment --coins BTC --period 24h --profile tradebot ``` **AI must output a structured market context summary before proceeding:** ``` [MARKET CONTEXT] Trend: Supertrend={UP|DOWN}, ADX={value} → {trending if >25 / ranging if <20} Volatility: BBWidth={value}, HV={value} → {high/medium/low} volatility regime Cap Flow: MFI={value} → {inflow if >50 / outflow if <50}, Funding={value} → {long-biased/short-biased/neutral} Positioning: Long={ratio}, Short={ratio} → {crowded longs/crowded shorts/balanced} Sentiment: BTC={bullish|neutral|bearish}, mentions={n} Baseline: Market is [description]. Overall confidence adjustment: [+1 / 0 / -1] ``` **Interpretation rules (AI applies judgment, not just thresholds):** - Supertrend DOWN + ADX > 25 = strong downtrend → penalize BTC long confidence by 2pts - MFI < 30 = capital outflow despite price drop → potential capitulation → boost long confidence 1pt - Funding rate persistently negative (< -0.01%) = shorts dominating, squeeze risk → boost long confidence 1pt - Long/short ratio < 0.45 (crowded shorts) = contrarian bullish signal → boost long confidence 1pt - BBWidth high + HV high = trending/breakout environment → good for directional trades, less ideal for grids - BBWidth low + HV low = compression → grid bots perform well, directional trades less reliable --- ## PHASE 0.6 — Gemma4 综合市场研判（每 cycle 一次，紧接 Phase 0.5 后） 在 Phase 0.5 并行执行的同时，额外获取多时间框架蜡烛： ```bash # 并行执行（与 Phase 0.5 同批） NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market candles BTC-USDT-SWAP --bar 1D --limit 7 --profile tradebot > /tmp/btc_1d.txt NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market candles BTC-USDT-SWAP --bar 12H --limit 7 --profile tradebot > /tmp/btc_12h.txt NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market candles BTC-USDT-SWAP --bar 4H --limit 12 --profile tradebot > /tmp/btc_4h.txt NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market candles BTC-USDT-SWAP --bar 1H --limit 12 --profile tradebot > /tmp/btc_1h.txt NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market candles BTC-USDT-SWAP --bar 5m --limit 20 --profile tradebot > /tmp/btc_5m.txt ``` 待 Phase 0.5 数据就绪后，构建综合 Prompt 调用本地 Gemma4（一次调用约 25 秒）： ```bash python3 -c " import sys c1d = open('/tmp/btc_1d.txt').read() c12h = open('/tmp/btc_12h.txt').read() c4h = open('/tmp/btc_4h.txt').read() c1h = open('/tmp/btc_1h.txt').read() # 注入 Phase 0.5 已知值 prompt = '''You are a senior crypto trading analyst. Analyze BTC-USDT-SWAP across all timeframes. [DAILY CANDLES - last 7, newest first] {c1d} [12H CANDLES - last 7, newest first] {c12h} [4H CANDLES - last 12, newest first] {c4h} [1H CANDLES - last 12, newest first] {c1h} [MARKET INDICATORS] Supertrend(1H):{st} ADX:{adx} DI+:{dip} DI-:{dim} MFI(1H):{mfi} Funding:{fr}% Long:{lr}% Short:{sr}% RSI(5m):{rsi} CCI(5m):{cci} Determine the macro trend from Daily/12H, then the intermediate trend from 4H/1H. Identify key support and resistance levels. Output EXACTLY 11 lines with no extra text: TREND_DAILY: UP or DOWN or SIDEWAYS TREND_12H: UP or DOWN or SIDEWAYS TREND_4H: UP or DOWN or SIDEWAYS TREND_1H: UP or DOWN or SIDEWAYS SUPPORT_1: [most critical support as integer] SUPPORT_2: [second support as integer] RESISTANCE_1: [most critical resistance as integer] RESISTANCE_2: [second resistance as integer] DIRECTION: LONG or SHORT or FLAT CONFIDENCE: HIGH or MEDIUM or LOW RISK: [main risk in 10 words max]'''.format(c1d=c1d,c12h=c12h,c4h=c4h,c1h=c1h,...) print(prompt) " | ollama run gemma4:e4b 2>/dev/null | col -b > /tmp/gemma4_out.txt ``` **解析输出并更新 state.gemma4_analysis（含新字段）：** ```python # 从 /tmp/gemma4_out.txt 解析 11 个字段 # 写入 state.json 的 gemma4_analysis: { "trend_daily": "DOWN", "trend_12h": "DOWN", "trend_4h": "DOWN", "trend_1h": "DOWN", "support_1": 75700, "support_2": 75000, "resistance_1": 78300, "resistance_2": 76100, "direction": "SHORT", # ← 核心方向判断 "confidence": "HIGH", "updated_at": "..." } ``` **必须输出：** ``` [GEMMA4 MACRO ANALYSIS {timestamp}] Daily={TREND} | 12H={TREND} | 4H={TREND} | 1H={TREND} SUPPORT_1={price} | SUPPORT_2={price} RESISTANCE_1={price} | RESISTANCE_2={price} DIRECTION={LONG/SHORT/FLAT} | CONFIDENCE={HIGH/MEDIUM/LOW} RISK: {text} → BTC交易方向: {做多/做空/观望} → 入场模式: {见下表} ``` **宏观方向决策（Gemma4 DIRECTION 字段）：** | Gemma4 DIRECTION | BTC 操作 | |------------------|---------| | LONG | 执行 Phase 2C 多单评分逻辑 | | SHORT | 执行 Phase 2C 空单评分逻辑（Gate/指标取反） | | FLAT | 跳过新 BTC 入场，仅管理现有持仓 | **入场门槛由宏观趋势一致性决定：** | Daily | 12H | 4H | 1H | 模式 | 满仓门槛 | 半仓门槛 | |-------|-----|----|----|------|----------|----------| | UP | UP | UP | UP/SW | 强顺势 | ≥65 | 50–64 | | UP | UP | DOWN | ANY | 回调入场 | ≥70 | 55–69 | | DOWN | DOWN | DOWN | UP | 逆势短反弹 | ≥82 | 67–81 | | DOWN | DOWN | DOWN | DOWN | 强逆势/做空 | LONG≥88 / SHORT≥70 | —/55–69 | --- ## PHASE 0.9 — Dynamic SL Algorithm (v3.14, 自适应) **来源**: Trail SL太紧 (BTC 15点buffer被chop洗) / RAVE SL 0.9%秒被洗 / BASED SL 6%被6.4%反弹打掉 ### SL距离公式 ```python def calc_sl_distance(instId, entry_price, direction): atr_1m = get_atr(instId, "1m", 14) atr_5m = get_atr(instId, "5m", 14) vol_24h = (24h_high - 24h_low) / avg * 100 # % # Base distance base_dist = max( 3 * atr_1m, # 避免1m chop 0.3% * entry_price, # 下限 2 * atr_5m # 避免5m noise ) # Volatility adjustment if vol_24h > 50: # 极端波动 (RAVE类) base_dist *= 3 # → 最少15-25% SL elif vol_24h > 30: # 高波动 (BASED类) base_dist *= 2 # → 最少8-15% SL elif vol_24h > 15: # 中波动 (PIPPIN类) base_dist *= 1.5 # else 标准 # Structure override (final check) nearest_swing = get_nearest_structure_point(direction) structural_dist = abs(entry_price - nearest_swing) + 0.3% * entry_price return max(base_dist, structural_dist) ``` ### 为不同品种的最低SL距离 | 标的类型 | 24h波动 | 最小SL% | |---|---|---| | BTC/ETH | <10% | 1-2% | | 主流altcoin | 10-20% | 3-5% | | 中波动 | 20-30% | 8-12% | | 高波动 | 30-50% | 15-25% | | 极端(RAVE类) | >50% | **≥25%** 必须 | ### Trail SL 距离 (profit保护) ``` Trail distance = max( 3 × 1m ATR, # 避免chop洗出 current_price × 0.5%, # 最小buffer 8% * entry_price # 高vol asset下限 ) ``` **铁律**: 永远不使用 **<2% price** SL (除非super低vol) --- ## PHASE 0.95 — Post-Fill OCO Guardian (v3.14, 每cycle必跑) **来源**: BTC OCO sz=4 vs position=9 有5张裸仓 bug ### 每cycle开始时必跑 ```bash # For each active position for POSITION in $(get_all_positions); do pos_sz=$(get_position_size $POSITION) oco_sz=$(get_oco_size $POSITION) if [ "$oco_sz" == "" ]; then # 无OCO, 立即创建 create_oco_from_memory_params $POSITION log "OCO GUARDIAN: Created OCO for ${POSITION} sz=${pos_sz}" elif [ "$oco_sz" != "$pos_sz" ]; then # OCO size不同步, amend amend_oco_size $POSITION $pos_sz log "OCO GUARDIAN: Amended ${POSITION} OCO sz ${oco_sz} → ${pos_sz}" else log "OCO GUARDIAN: ${POSITION} OK sz=${pos_sz}" fi done ``` ### 触发场景 - 分层limit filled → 自动sync - 手动加market order → 自动sync - OCO被手动cancel → 自动recreate --- ## PHASE 1 — Account Check (every cycle, run first) ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx account balance --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx account positions --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx bot grid orders --profile tradebot ``` Extract: - `free_usdt`: available USDT in trading account - `btc_pos`: BTC-USDT-SWAP position size, side, avg entry price, uPnL - `rave_pos`: RAVE-USDT-SWAP position size, side, avg entry price, uPnL - `active_grids`: list of running grid bots with algoId and instId Reconcile with state.json. If a position exists on exchange but not in state, add it. If state shows active but exchange shows closed, clear it from state. --- ## PHASE 2 — BTC Strategy (400 USDT, 50x isolated) ### Step 2A — One-Time Activation Check If `state.btc_activated == false`: ```bash # Get 1H MACD for BTC NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator MACD BTC-USDT-SWAP --bar 1H --profile tradebot ``` Parse the DIF line (fast line) value from the response. - If `abs(DIF) <= 50` (DIF is near zero): set `state.btc_activated = true`, proceed to Step 2B - Otherwise: **skip entire Phase 2 this cycle**. Log: "BTC strategy not yet activated. 1H DIF = {value}, waiting for zero-cross." If `state.btc_activated == true`: always proceed to Step 2B. ### Step 2B — Gather BTC Signals (multi-dimensional) ```bash # Price momentum NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator RSI BTC-USDT-SWAP --period 14 --bar 5m --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator CCI BTC-USDT-SWAP --period 20 --bar 5m --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market ticker BTC-USDT-SWAP --profile tradebot # Volume-weighted capital flow NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator mfi BTC-USDT-SWAP --bar 5m --profile tradebot # Candlestick reversal patterns (CLI) NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator bull-engulf BTC-USDT-SWAP --bar 5m --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator three-soldiers BTC-USDT-SWAP --bar 5m --profile tradebot # 30m candles for trend filter (MACD uses 1H - 30m bar not supported by OKX MACD endpoint) NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market candles BTC-USDT-SWAP --bar 30m --limit 22 --profile tradebot > /tmp/btc_30m.txt NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator MACD BTC-USDT-SWAP --bar 1H --profile tradebot > /tmp/btc_1h_macd.txt # 5m candles for pattern analysis NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market candles BTC-USDT-SWAP --bar 5m --limit 20 --profile tradebot > /tmp/btc_5m.txt ``` **Step 2B.5 — 30m Trend Filter（逆势判断）** 从 `/tmp/btc_30m.txt` 取最近20根30m收盘价，计算EMA20： ```python k = 2 / (20 + 1) ema = closes[0] for c in closes[1:]: ema = c * k + ema * (1 - k) ``` ``` if btc_price < ema20_30m: → 逆势模式 (counter-trend): 满仓门槛提升至 ≥80，半仓门槛提升至 65-79 else: → 顺势模式 (with-trend): 保持标准门槛 ≥70 / 50-69 输出: [TREND FILTER] 30m EMA20={ema:.1f}, price={price}, mode={逆势/顺势} ``` **Step 2B.6a — Pinbar 自动检测（从K线数据直接计算，无需AI）** ```bash python3 -c " import re data = open('/tmp/btc_5m.txt').read() # OKX candle format: "2026/4/19 02:05:00 open high low close vol" # parts[0]=date parts[1]=time parts[2]=open parts[3]=high parts[4]=low parts[5]=close lines = [l.strip() for l in data.strip().split('\n') if re.match(r'2026', l.strip())] if lines: parts = lines[0].split() try: o,h,l,c = float(parts[2]),float(parts[3]),float(parts[4]),float(parts[5]) body = abs(c - o) total = h - l lower = min(o,c) - l upper = h - max(o,c) if total > 0: if lower >= 2*body and body <= total/3.0 and lower > upper: print('PINBAR: BULLISH lower={:.2f} body={:.2f} ratio={:.2f}'.format(lower,body,lower/max(body,0.01))) elif upper >= 2*body and body <= total/3.0 and upper > lower: print('PINBAR: BEARISH upper={:.2f} body={:.2f} ratio={:.2f}'.format(upper,body,upper/max(body,0.01))) else: print('PINBAR: none') else: print('PINBAR: none') except: print('PINBAR: none') else: print('PINBAR: none') " > /tmp/btc_pinbar.txt cat /tmp/btc_pinbar.txt ``` 解析规则（比较当前价格与 gemma4_analysis.support_1）： - `PINBAR: BULLISH` 且当前价格在 `support_1 * 0.985 ~ support_1 * 1.015`（支撑位±1.5%） → **关键支撑位Pinbar**，高价值信号 - `PINBAR: BULLISH` 但不在支撑位附近 → 普通Pinbar，参考价值 - `PINBAR: BEARISH` → 做多时为负面信号 - `PINBAR: none` → 无Pinbar **Step 2B.6b — Gemma4 扩展K线形态识别（本地AI辅助，与2B.6a互补）** 在Pinbar检测之后，用本地Gemma4识别更复杂的多K线形态（晨星、穿刺线等）： ```bash echo "BTC-USDT-SWAP 5m OHLC candles (newest first, format: time open high low close vol): $(cat /tmp/btc_5m.txt) Task: Identify multi-candle bullish reversal patterns ONLY (morning star, bullish harami, piercing line, tweezer bottom, three white soldiers). Single-candle patterns are already handled. Reply with exactly one line: PATTERN: [pattern_name] or PATTERN: none" \ | ollama run gemma4:e4b 2>/dev/null \ | col -b | grep "PATTERN:" | tail -1 > /tmp/gemma4_pattern.txt cat /tmp/gemma4_pattern.txt ``` 解析规则： - 输出含 `PATTERN: none` 或文件为空 → Gemma4无多K线形态 - 输出含具体形态名 → Gemma4检测到多K线形态 ⚠️ 若ollama未启动，跳过2B.6b，仅使用2B.6a Pinbar结果。响应通常需要20-30秒。 **Step 2B.6c — 1H MACD 顶背离检测（持仓风险监控）** 注意：OKX MACD指标端点不支持30m时间框架，改用1H（`/tmp/btc_1h_macd.txt`）。 仅当 `state.btc_position.active == true` 时执行： ```bash python3 -c " import re # 解析 /tmp/btc_1h_macd.txt 获取最近10个DIF值（1H MACD，OKX不支持30m） data = open('/tmp/btc_1h_macd.txt').read() # 假设格式包含 dif 字段，提取最近值 difs = re.findall(r'dif[\":\s]+(-?[\d.]+)', data, re.IGNORECASE) difs = [float(x) for x in difs[:10]] # 解析 /tmp/btc_30m.txt 获取最近10根收盘价 price_data = open('/tmp/btc_30m.txt').read() price_lines = [l.strip() for l in price_data.strip().split('\n') if re.match(r'[0-9]', l.strip())] closes = [float(l.split()[4]) for l in price_lines[:10] if len(l.split()) >= 5] if len(difs) >= 6 and len(closes) >= 6: # 找最近的两个价格高点（swing high） swing_highs = [] for i in range(1, len(closes)-1): if closes[i] > closes[i-1] and closes[i] > closes[i+1]: swing_highs.append((i, closes[i], difs[i] if i < len(difs) else None)) if len(swing_highs) >= 2: sh1, sh2 = swing_highs[-2], swing_highs[-1] # sh2 is more recent if sh2[2] is not None and sh1[2] is not None: if sh2[1] > sh1[1] and sh2[2] < sh1[2]: # Price HH but MACD LH print('MACD_DIV: BEARISH price_sh1={:.0f} price_sh2={:.0f} dif_sh1={:.2f} dif_sh2={:.2f}'.format(sh1[1],sh2[1],sh1[2],sh2[2])) else: print('MACD_DIV: none') else: print('MACD_DIV: insufficient_dif_data') else: print('MACD_DIV: insufficient_swings') else: print('MACD_DIV: insufficient_data') " > /tmp/btc_macd_div.txt cat /tmp/btc_macd_div.txt ``` ### Step 2C — BTC Entry Weighted Scoring (total 100 pts) **Only if no active BTC position** (`state.btc_position.active == false`): **Gate condition** (must pass before scoring): `rsi_5m < 45` AND `cci_5m < -60` Gate fails → output skip reason and stop. Gate passes → proceed to full scoring. **趋势门槛调整（来自Step 2B.5的逆势/顺势模式）：** | 模式 | 满仓门槛 | 半仓门槛 | 说明 | |------|----------|----------|------| | 顺势 (btc_price ≥ 30m EMA20) | ≥70 | 50–69 | 标准门槛 | | 逆势 (btc_price < 30m EMA20) | **≥80** | **65–79** | 下跌趋势中抄底需更强信号 | **Weighted scoring table (8 dimensions, 100 pts total):** | # | Dimension | Weight | Scoring Rules | |---|-----------|--------|---------------| | 1 | RSI 5m (momentum depth) | **15 pts** | <30→15 / <35→12 / <40→9 / <45→5 / ≥45→0 | | 2 | CCI 5m (oscillator depth) | **15 pts** | <-150→15 / <-100→12 / <-80→8 / <-60→4 / ≥-60→0 | | 3 | MFI 5m (capital outflow) | **15 pts** | <30→15 / <40→11 / <50→6 / ≥50→0 | | 4 | Supertrend 1H (trend) | **15 pts** | UP→15 / DOWN→0 | | 5 | Long/Short ratio (positioning) | **10 pts** | short>65%→10 / short>55%→6 / balanced→3 / long>55%→0 | | 6 | Funding rate (market bias) | **10 pts** | <-0.02%→10 / <-0.01%→7 / ±0.01%→4 / >0.02%→0 | | 7 | K线形态+热点共振 | **10 pts** | Pinbar在support_1±1.5%→10 / CLI(bull-engulf/three-soldiers)→8 / Pinbar非关键位→6 / Gemma4多K线形态→6 / 价格在support_1±0.5%且30mEMA20±0.5%热点共振→+2叠加(上限10) / 全无→0 | | 8 | Gemma4综合判断 | **10 pts** | BIAS=BULLISH+HIGH→10 / BULLISH+MEDIUM→8 / NEUTRAL+HIGH→5 / NEUTRAL+MEDIUM→3 / BEARISH→0 | **Total = sum of all 8 dimensions = max 100 pts** **AI must output the scoring block:** ``` [BTC入场加权评分] ① RSI={值} → {分}/15分 ② CCI={值} → {分}/15分 ③ MFI={值} → {分}/15分 ④ Supertrend={方向} → {分}/15分 ⑤ 多空比=多{%}/空{%} → {分}/10分 ⑥ 资金费率={值} → {分}/10分 ⑦ K线形态={结果} → {分}/10分 ⑧ Gemma4={BIAS/CONF} → {分}/10分 ────────────────────────────── 总分: {分}/100分 入场模式: {强顺势/震荡反弹/短线反弹/强逆势}（来自Phase 0.6） 决策: {满仓/半仓/跳过}（门槛见 Phase 0.6 表格） 保证金: {200U / 100U / 0U} 推理: [2-3句：关键支撑维度，拖分维度，最大风险] ``` **Execution — 方向由 Phase 0.6 Gemma4 DIRECTION 决定：** **做多（DIRECTION=LONG）：** Gate: rsi_5m < 45 AND cci_5m < -60 ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap leverage --instId BTC-USDT-SWAP --lever 50 --mgnMode isolated --posSide long --profile tradebot # 验证: okx swap get-leverage --instId BTC-USDT-SWAP --mgnMode isolated --profile tradebot okx swap place --instId BTC-USDT-SWAP --side buy --ordType market \ --sz {sz} --posSide long --tdMode isolated --tag agentTradeKit --profile tradebot ``` **做空（DIRECTION=SHORT）：** Gate: rsi_5m > 55 AND cci_5m > 60 - 评分规则取反：RSI>70→15 / RSI>65→12 / RSI>60→9 / RSI>55→5 - CCI>150→15 / CCI>100→12 / CCI>80→8 / CCI>60→4 - MFI>70→15 / MFI>60→11 / MFI>50→6 - Supertrend DOWN→15 / UP→0 - K线形态：bear-engulf/three-ravens=1 → 10pts（CLI反向形态） - Gemma4 DIRECTION=SHORT+CONFIDENCE=HIGH → 10pts ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap leverage --instId BTC-USDT-SWAP --lever 50 --mgnMode isolated --posSide short --profile tradebot okx swap place --instId BTC-USDT-SWAP --side sell --ordType market \ --sz {sz} --posSide short --tdMode isolated --tag agentTradeKit --profile tradebot ``` 空单止损：`sl_price = avg_entry * 1.012`（价格上涨1.2%止损） 空单止盈：TP① RSI<30 OR CCI<-80 减仓20%；TP② RSI<25 全平 门槛仍按宏观趋势一致性表格执行（见 Phase 0.6）。 **DIRECTION=FLAT**：跳过所有新 BTC 入场。 ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap set-leverage --instId BTC-USDT-SWAP --lever 50 --mgnMode isolated --posSide long --profile tradebot ``` ``` sz = floor(margin * 50 / (btc_price * 0.01)) ``` ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap place --instId BTC-USDT-SWAP --side buy --ordType market \ --sz {sz} --posSide long --tdMode isolated --tag agentTradeKit --profile tradebot ``` Update state: active=true, entry_count=1, avg_entry_price, total_sz, remaining_pct=100. **Second entry** (entry_count==1, next signal with score ≥ 50): place same sz, update weighted avg. **入场后立即挂 OCO 原生止损止盈（OKX 平台硬止损，断网/宕机也会执行）：** ``` sl_price = avg_entry_price * 0.988 # 入场价 -1.2% tp_price = state.gemma4_analysis.resistance_1 # Gemma4 最近阻力位 ``` ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap algo place \ --instId BTC-USDT-SWAP --side sell --sz {total_sz} --ordType oco \ --tpTriggerPx {tp_price} --tpOrdPx=-1 \ --slTriggerPx {sl_price} --slOrdPx=-1 \ --posSide long --tdMode isolated --reduceOnly \ --tag agentTradeKit --profile tradebot ``` 将返回的 `algoId` 存入 state: `"btc_oco_algoId": "{algoId}"`。 OCO 触发后自动全平12张，无需 Step 2D 手动止损。 ### Step 2D — BTC Exit Logic **Only if active BTC position** (`state.btc_position.active == true`): ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator RSI BTC-USDT-SWAP --period 14 --bar 5m --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator CCI BTC-USDT-SWAP --period 20 --bar 5m --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market ticker BTC-USDT-SWAP --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap algo orders --instId BTC-USDT-SWAP --profile tradebot ``` **⚡ OCO 原生止损优先（OKX 平台级，断线也生效）:** 首先检查 `state.btc_oco_algoId` 是否存在且algo orders中仍有效： - 若 OCO 已触发（orders为空 + 仓位消失）→ 清除 state btc_position，更新 peak_equity，跳过后续手动逻辑 - 若 OCO 存在但 Gemma4 更新了 resistance_1 或 support_1 → 用 `okx swap algo amend` 更新触发价 若 OCO 不存在（遗漏或已取消），立即补挂： ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap algo place \ --instId BTC-USDT-SWAP --side sell --sz {total_sz} --ordType oco \ --tpTriggerPx {resistance_1} --tpOrdPx=-1 \ --slTriggerPx {avg_entry * 0.988} --slOrdPx=-1 \ --posSide long --tdMode isolated --reduceOnly \ --tag agentTradeKit --profile tradebot ``` **Stop Loss B（备用，当 OCO 意外失效时）— Gemma4 支撑位跌破止损:** ``` support_1 = state.gemma4_analysis.support_1 if btc_price <= support_1 AND state.gemma4_analysis.trend_1h == "DOWN": → 取消 OCO：okx swap algo cancel --instId BTC-USDT-SWAP --algoId {btc_oco_algoId} → 立即全平（支撑位破位，下跌加速风险） → 日志: "SUPPORT BREAK STOP: price={price} broke support_1={support_1}" ``` 若手动触发全平： ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap place --instId BTC-USDT-SWAP --side sell --ordType market \ --sz {state.btc_position.total_sz} --posSide long --tdMode isolated --reduceOnly \ --tag agentTradeKit --profile tradebot ``` Clear btc_position and btc_oco_algoId from state. **TP0 — 价格过度延伸（均值回归，优先级高于常规TP）:** ``` 获取30m EMA20值（来自/tmp/btc_30m.txt的最新收盘价序列计算，或从已有指标估算）： if btc_price > ema20_30m * 1.030: → 价格偏离30mEMA20超过3%，均值回归概率高 → 触发TP① (减仓20%)，日志: "OVEREXTENSION TP0: price={price} > EMA20*1.03={threshold}" ``` **TP1.5 — MACD 1H 顶背离（动量衰竭信号，优先级高于常规TP①）:** ``` 读取 /tmp/btc_macd_div.txt： if contains "MACD_DIV: BEARISH": → 1H MACD顶背离，价格创新高但MACD动量下降，上涨动力衰竭 → 触发TP① (减仓20%)，日志: "MACD_DIV TP1.5: bearish divergence detected on 1H" ``` **动态止盈阈值（根据 Gemma4 BIAS 自动调整）：** | Gemma4 BIAS | TP① 触发 | TP② 触发 | TP 目标参考 | |-------------|---------|---------|-----------| | BULLISH | CCI > 100 | RSI > 70 | 标准（顺势持有更久）| | NEUTRAL | CCI > 90 | RSI > 67 | 略保守 | | BEARISH | CCI > 80 | RSI > 63 | 保守（反弹随时结束）| BEARISH 时额外规则：若价格触及 `state.gemma4_analysis.resistance_2` → 视同 TP① 触发（取决于哪个先发生）。 **Take Profit ① — CCI > {动态阈值}** (if `remaining_pct == 100`): AI must output reasoning before executing: ``` [BTC TP① 推理] CCI={值} 超买（当前阈值={动态阈值}，BIAS={BIAS}）。未实现盈亏={uPnL}U。 评估：{反弹幅度是否接近RESISTANCE_2；BEARISH模式下应保守减仓} 决策: 执行TP① 减仓20% ``` ``` close_sz = floor(state.btc_position.total_sz * 0.20) ``` ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap place --instId BTC-USDT-SWAP --side sell --ordType market \ --sz {close_sz} --posSide long --tdMode isolated --reduceOnly \ --tag agentTradeKit --profile tradebot ``` Update `remaining_pct = 80`. **Take Profit ② — RSI > {动态阈值}** (close ALL remaining): AI must output reasoning before executing: ``` [BTC TP② 推理] RSI={值} 超买（当前阈值={动态阈值}，BIAS={BIAS}）。 持仓均价={avg}，当前价={price}，盈利={profit_pct:.1f}%。 评估：{若BEARISH模式，当前是否接近resistance_1；综合判断出场合理性} 决策: 执行TP② 平全部剩余仓位 ``` ``` remaining_sz = floor(state.btc_position.total_sz * remaining_pct / 100) ``` ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap place --instId BTC-USDT-SWAP --side sell --ordType market \ --sz {remaining_sz} --posSide long --tdMode isolated --reduceOnly \ --tag agentTradeKit --profile tradebot ``` Clear btc_position from state. --- ## PHASE 2.6 — RAVE Wick Hunter (v3.14, 高波动insta-spike狙击) **Motivation**: 今日RAVE多次秒级插针 (+5-10%单根K线), 本应重仓做空但我错过。此模块**专门**捕捉插针做空。 ### Trigger 条件 (AI综合评估) 当 **RAVE当前价 < 1.30** (处于下跌结构末期) 且 **触发条件任一**: **T1 — 大上影wick**: - 最近1m K线: `(high - max(open,close)) / close > 3%` - 即上影>3%且收盘回到实体 **T2 — 单根intrabar spike**: - 最近3根1m任一: `(high - low)/open > 5%` - 说明market内正在剧烈波动 **T3 — 近期大阻力wick到**: - 当前价 touch近期 (30min内) 重要阻力 (H2/H3/5m BB upper) 但未收盘突破 - 典型"fake breakout"模式 ### AI Decision Layer (必须, 比赛合规) 每次trigger发生, AI评估: ``` [RAVE WICK HUNTER {时间}] Input: - Trigger type: T1/T2/T3 - Current price: {price} - 1m bar pattern: {OHLC最近5根} - 5m Supertrend: {UP/DOWN} - 15m Supertrend: {UP/DOWN} - 1H Supertrend: {UP/DOWN} - Float PnL全仓: {amount} - Account health: {equity} - 距24h low: {pct} AI Synthesis (必须文字): 评估1: 这是"真rejection"还是"假wick后趋势反转"? 评估2: 重仓做空 vs 试探仓? 评估3: 入场方式 — 市价 (wick正在发生) vs 限价 (等next wick)? 评估4: 当前全仓暴露, 加空否? Final decision: {GO_HEAVY | GO_SMALL | SKIP} Position sizing: - GO_HEAVY: 40U margin × 5x = $200 notional (≈150-160 RAVE contracts) - GO_SMALL: 20U margin × 5x = $100 notional Entry: - 若spike正在发生 → market short 立即 - 若刚rejected → limit at 95% of wick high (等小反弹) SL: wick的最高价 × 1.12 (12% buffer, 符合v3.14高vol规则) TP: 24h low × 1.02 (大目标, 激进) Confidence: {0-1} Risk assessment: 若SL触发最大损失 {amount} ``` ### 限价埋伏 (Pre-Placed Wick Traps) 预先挂单在可能的wick高位, 实现**零延迟捕捉**: ``` For RAVE (current ~1.20): - 限价SHORT @ 1.35 (近期wick peak) - 限价SHORT @ 1.50 (大wick target) - 限价SHORT @ 1.75 (极端wick, 抓一次就赚大) 每个单独OCO: TP: 1.00 (统一大目标) SL: entry × 1.15 (符合高vol rule) sz: 5-10张 (每单15-30U margin) ``` **优势**: 限价不fill=零风险; 只要wick触及, 立即成交并带OCO保护。 ### 风控 - 本Phase最大total allocation: 80U margin (即便3层全fill) - 24h RAVE波动>50% 则启用此phase - RAVE持仓 + wick hunter 总仓位max 150U margin - 若equity < 550U, 停止Wick Hunter (保护剩余资金) --- ## PHASE 2.5 — RAVE Dedicated Short (up to 200 USDT, 10x isolated) **Budget rule**: If `state.btc_position.active == false`, RAVE may use up to 200 USDT. Otherwise max 100 USDT. ### Step 2.5A — Multi-Timeframe Bearish Signal Detection Gather RAVE data across 4 timeframes: ```bash # 4H signals NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator RSI RAVE-USDT-SWAP --period 14 --bar 4H --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator MACD RAVE-USDT-SWAP --bar 4H --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market candles RAVE-USDT-SWAP --bar 4H --limit 20 --profile tradebot # 1H signals NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator RSI RAVE-USDT-SWAP --period 14 --bar 1H --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator MACD RAVE-USDT-SWAP --bar 1H --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market candles RAVE-USDT-SWAP --bar 1H --limit 20 --profile tradebot # 5m signals NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator RSI RAVE-USDT-SWAP --period 14 --bar 5m --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator MACD RAVE-USDT-SWAP --bar 5m --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market candles RAVE-USDT-SWAP --bar 5m --limit 20 --profile tradebot # Current price NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market ticker RAVE-USDT-SWAP --profile tradebot ``` **Bearish signals to detect** (any ONE is sufficient to trigger initial entry): 1. **RSI Bearish Divergence (any TF)**: On the last 20 candles, find two swing highs. If price swing_high_2 > swing_high_1 (higher high) but RSI at swing_high_2 < RSI at swing_high_1 (lower RSI) → bearish divergence confirmed. 2. **MACD Death Cross (any TF)**: DIF crosses below DEA on the most recent candle → death cross signal. 3. **MACD Bearish Divergence (any TF)**: Two DIF/DEA crossover points. If crossover_2_DIF < crossover_1_DIF but price made higher high → MACD bearish divergence. 4. **Top Fractal (顶分型)**: In 4H or 1H candles, find a 3-candle pattern where middle candle has the highest high → fractal top formed. 5. **RSI Overbought (any TF)**: RSI > 70 after a major rally → potential reversal zone. 6. **MACD Head-and-Shoulders Top (4H/1H)**: In the last 20 candles of 4H or 1H MACD, find 3 DIF peaks where peak1 < peak2 (left shoulder < head) AND peak3 < peak2 (right shoulder < head). This signals Elliott Wave 5 completion / exhaustion top. A right shoulder where DIF failed to reach the head level = distribution phase ending → strong short signal. Extra weight: if peak3 < peak1 also (classic 头肩顶), add +1 to signal_count. Count total number of bearish signals found across all timeframes: `signal_count`. **Identify swing_high**: Find the highest price across the last 20 candles in the signal-triggering timeframe. This is the stop-loss reference level. ### Step 2.5B — RAVE Entry Logic **If `state.rave_position.active == false` AND `signal_count >= 1`**: ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap set-leverage --instId RAVE-USDT-SWAP --lever 10 --mgnMode isolated --posSide short --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market instruments --instType SWAP --instId RAVE-USDT-SWAP --profile tradebot ``` Initial entry = 40 USDT margin: ``` notional = 40 * 10 = 400 USDT sz = floor(notional / rave_price) — use ctVal from instruments response ``` ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap place --instId RAVE-USDT-SWAP --side sell --ordType market \ --sz {sz} --posSide short --tdMode isolated --tag agentTradeKit --profile tradebot ``` Update state: ```json { "rave_position": { "active": true, "avg_entry_price": rave_price, "total_sz": sz, "remaining_pct": 100, "signal_count": signal_count, "swing_high": swing_high_price } } ``` **If `state.rave_position.active == true` AND new signals detected** (signal_count > state.rave_position.signal_count): Add-on position (scale in) with additional margin based on signal count: - 2 signals: add 40 USDT margin - 3+ signals: add 60 USDT margin (total up to budget limit) Check budget limit first: ``` current_margin = state.rave_position.total_sz * state.rave_position.avg_entry_price / 10 addon_allowed = rave_budget - current_margin ``` If `addon_allowed <= 0`: skip add-on. ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap place --instId RAVE-USDT-SWAP --side sell --ordType market \ --sz {addon_sz} --posSide short --tdMode isolated --tag agentTradeKit --profile tradebot ``` Recalculate avg_entry_price (weighted average). Update signal_count and total_sz. ### Step 2.5B2 — Keltner Channel Rebound Add-on (5m, replaces Vegas Channel) **Only if `state.rave_position.active == true`** AND budget has room AND `state.rave_position.vegas_addon_count < 3`: ```bash # 5m Keltner Channel (EMA-based, similar to Vegas Channel logic) NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator keltner RAVE-USDT-SWAP --bar 5m --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market ticker RAVE-USDT-SWAP --profile tradebot ``` Parse: `upper`, `middle`, `lower` from Keltner response. **Keltner Middle Rebound condition** (all must be true): 1. `rave_price >= middle * 0.99` AND `rave_price <= middle * 1.01` — price touching middle band (±1%) 2. `rave_price < state.rave_position.avg_entry_price` — still below entry (profit zone) 3. Budget has room: `current_margin < rave_budget_limit` **AI Reasoning**: Current price={rave_price}, Keltner middle={middle}, upper={upper}, lower={lower}. Is price rebounding into the Keltner middle as resistance? If all conditions met — add-on 40 USDT margin: ``` sz = floor(40 * effective_leverage / (rave_price * ctVal)) # RAVE effective lever = 3 ``` ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap place --instId RAVE-USDT-SWAP --side sell --ordType market \ --sz {sz} --posSide short --tdMode isolated --tag agentTradeKit --profile tradebot ``` Recalculate avg_entry_price (weighted average). Update total_sz. Increment `state.rave_position.vegas_addon_count` by 1. Note: vegas_addon_count resets to 0 when rave_position is cleared (TP or SL). ### Step 2.5B3 — VWAP Rebound Add-on (5m, replaces BB middle band) **Only if `state.rave_position.active == true`** AND budget has room AND `state.rave_position.bb_addon_count < 3`: ```bash # 5m VWAP as dynamic resistance reference NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator vwap RAVE-USDT-SWAP --bar 5m --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market ticker RAVE-USDT-SWAP --profile tradebot ``` Parse: `vwap` value from response. **VWAP Rebound condition** (all must be true): 1. `rave_price >= vwap * 0.99` AND `rave_price <= vwap * 1.01` — price touching VWAP (±1%) 2. `rave_price < state.rave_position.avg_entry_price` — still below entry (profit zone) 3. Budget has room: `current_margin < rave_budget_limit` **AI Reasoning**: Current price={rave_price}, VWAP={vwap}. Is price rebounding up to VWAP resistance from below? If all conditions met — add-on 40 USDT margin: ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap place --instId RAVE-USDT-SWAP --side sell --ordType market \ --sz {sz} --posSide short --tdMode isolated --tag agentTradeKit --profile tradebot ``` Recalculate avg_entry_price. Update total_sz. Increment `state.rave_position.bb_addon_count` by 1. Note: bb_addon_count resets to 0 when rave_position is cleared. ### Step 2.5C — RAVE Exit Logic **Only if `state.rave_position.active == true`**: ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market ticker RAVE-USDT-SWAP --profile tradebot ``` **Stop Loss — highest priority**: If `rave_price >= state.rave_position.swing_high`: ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap place --instId RAVE-USDT-SWAP --side buy --ordType market \ --sz {state.rave_position.total_sz} --posSide short --tdMode isolated --reduceOnly \ --tag agentTradeKit --profile tradebot ``` Clear rave_position from state. Log: "RAVE STOP LOSS: price {rave_price} broke above swing_high {swing_high}" **Take Profit calculation**: ``` drop_pct = (state.rave_position.avg_entry_price - rave_price) / state.rave_position.avg_entry_price * 100 ``` **TP① — drop >= 48%** (if `remaining_pct == 100`): ``` close_sz = max(1, floor(state.rave_position.total_sz * 0.50)) ``` Close 50% (at least 1 contract). Update `remaining_pct = 50`. ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap place --instId RAVE-USDT-SWAP --side buy --ordType market \ --sz {close_sz} --posSide short --tdMode isolated --reduceOnly \ --tag agentTradeKit --profile tradebot ``` Log: "RAVE TP1: closed 50% at {rave_price} (-{drop_pct:.1f}%)" **TP② — drop >= 70%** (if `remaining_pct == 50`): ``` remaining_sz = max(1, floor(state.rave_position.total_sz * remaining_pct / 100)) ``` Close all remaining. Clear rave_position from state. ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap place --instId RAVE-USDT-SWAP --side buy --ordType market \ --sz {remaining_sz} --posSide short --tdMode isolated --reduceOnly \ --tag agentTradeKit --profile tradebot ``` Log: "RAVE TP2: fully closed at {rave_price} (-{drop_pct:.1f}%)" **TP③ — drop >= 90%** (safety net, close all if still open): ``` remaining_sz = floor(state.rave_position.total_sz * remaining_pct / 100) ``` ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap place --instId RAVE-USDT-SWAP --side buy --ordType market \ --sz {remaining_sz} --posSide short --tdMode isolated --reduceOnly \ --tag agentTradeKit --profile tradebot ``` Clear rave_position from state. Log: "RAVE TP3: fully closed at {rave_price} (-{drop_pct:.1f}%)" --- ## PHASE 3 — High-Volatility Grid Strategy (100 USDT total, 10x isolated) ### Step 3A — Dual Candidate Scan Run both scans in parallel: **Scan A — Price Gainers:** ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market tickers SWAP --profile tradebot ``` Sort by 24h change descending. Take top 10 (exclude RAVE). For each, call `okx market ticker {instId}` to get exact 24h change %. **Scan B — OI Accumulation (capital flow signal):** ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market oi-change --instType SWAP --bar 1H --sortBy oiDeltaPct --sortOrder desc --limit 10 --minVolUsd24h 500000 --profile tradebot ``` Returns tokens where large positions are being built in the last hour. **Merge:** Tokens appearing in BOTH lists (price up AND OI up) are highest conviction. Single-list tokens are lower priority. Skip any already in `state.grid_bots`. ### Step 3B — Multi-Dimensional Candidate Validation For each candidate token (process top 3 from merged scan): ```bash # Technical: divergence detection NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator RSI {instId} --period 14 --bar 5m --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator MACD {instId} --bar 5m --profile tradebot # Volatility: is this a good grid environment? NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator bbwidth {instId} --bar 1H --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator atr {instId} --bar 5m --profile tradebot # Capital flow: confirm with MFI NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator mfi {instId} --bar 1H --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market funding-rate {instId} --profile tradebot # Sentiment: is the market aware of this move? NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx news coin-sentiment --coins {baseCcy} --period 24h --profile tradebot ``` **Divergence detection:** - **Bullish (底背离)** → LONG grid: price lower low + RSI higher low - **Bearish (顶背离)** → SHORT grid: price higher high + RSI lower high - **Neutral (no divergence)** → NEUTRAL grid if BBWidth confirms oscillation **Grid Candidate Weighted Scoring (total 100 pts):** | # | Dimension | Weight | Scoring Rules | |---|-----------|--------|---------------| | 1 | 来源扫描 (signal source) | **20 pts** | 价格+OI双榜→20 / 仅OI榜→12 / 仅价格榜→8 | | 2 | RSI背离 (divergence) | **20 pts** | 确认背离→20 / 弱背离→10 / 无→0 | | 3 | MACD方向确认 | **15 pts** | 与方向一致→15 / 中性→5 / 反向→0 | | 4 | MFI 1H (资金流向) | **15 pts** | >70→15 / 50-70→10 / 30-50→5 / <30→0 | | 5 | BBWidth (网格适配度) | **10 pts** | 低/中震荡环境→10 / 强趋势→4 | | 6 | 情绪 (sentiment) | **10 pts** | 看多>50%且热度>200→10 / 看多>40%且热度>100→6 / 中性→3 / 看空→0 | | 7 | 资金费率 (funding) | **10 pts** | 正常±0.02%→10 / 极端正>0.05%→3 / 极端负<-0.05%→3 | > **⚠️ 资金费率硬性拦截**：若 `abs(fundingRate) > 0.01`（即 >1%），**无论评分多高，强制跳过该标的**。 > 原因：1%/次 = 每天最高3%的资金成本，将快速侵蚀网格利润，不值得部署。 **AI must output for each candidate:** ``` [网格候选加权评分: {instId}] ① 来源={双榜/仅价格/仅OI} → {分}/20分 ② RSI背离={确认/弱/无} → {分}/20分 ③ MACD={金叉/死叉/中性} → {分}/15分 ④ MFI(1H)={值} → {分}/15分 ⑤ BBWidth={值}→{震荡/趋势} → {分}/10分 ⑥ 情绪={看多%}，{mentions}条 → {分}/10分 ⑦ 资金费率={值} → {分}/10分 ────────────────────────────── 总分: {分}/100分 决策: {部署网格≥70 / 减量部署50-69 / 少量试仓30-49 / 跳过<30} 网格方向: {做多/做空/中性} 仓位: {按评分区间，见下表} 推理: [2-3句：哪些维度是核心驱动，信号之间是否有矛盾， 盈利逻辑是什么，最大风险在哪里] ``` ### Step 3C — Position Sizing by Score ``` 预算检查：existing_grid_total + new_size ≤ 100 USDT，否则跳过。 按总分决定仓位： ≥ 70分 → min(remaining_budget, 70) USDT (高置信度) 50–69分 → min(remaining_budget, 40) USDT (中等置信度) 30–49分 → min(remaining_budget, 20) USDT (低置信度，试仓) < 30分 → 跳过 ``` ### Step 3D — Grid Bot Creation ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market indicator ATR {instId} --period 14 --bar 5m --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market ticker {instId} --profile tradebot ``` ``` grid_min = current_price - (1.5 * atr) grid_max = current_price + (1.5 * atr) # Stop loss: outside grid range with 10% buffer stop_loss_px = grid_min * 0.90 (for long/neutral grid) stop_loss_px = grid_max * 1.10 (for short grid) ``` ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx swap set-leverage --instId {instId} --lever 10 --mgnMode isolated --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx bot grid create \ --instId {instId} \ --algoOrdType contract_grid \ --minPx {grid_min} \ --maxPx {grid_max} \ --gridNum 10 \ --direction {long|short} \ --lever 10 \ --sz {position_size} \ --tag agentTradeKit \ --profile tradebot ``` Save to state.grid_bots: algoId, instId, direction, maxPx={grid_max}, minPx={grid_min}, sl_px={stop_loss_px}, lever=10, sz={position_size}. ### Step 3E — Grid Bot Management (existing bots) ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx bot grid details --algoId {algoId} --algoOrdType contract_grid --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market ticker {instId} --profile tradebot NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx market funding-rate {instId} --profile tradebot ``` **⚠️ 资金费率规避（最高优先级，先于止损检查）**： ``` now = current_time minutes_to_funding = (fundingTime - now) in minutes if abs(fundingRate) > 0.01 AND minutes_to_funding <= 30: → 立即停止网格以规避本次资金费 → 记录到state：paused_grid = {instId, algoId, direction, maxPx, minPx, sl_px, sz, resume_after: fundingTime + 10min} → 从state.grid_bots移除 → 执行: okx bot grid stop --algoId {algoId} --algoOrdType contract_grid --profile tradebot → 日志: "FUNDING AVOID: 停止 {instId} 网格，资金费率={fundingRate:.3%}，结算时间={fundingTime}" if instId in state.paused_grids AND now > resume_after: → 重建网格（使用原参数：direction, maxPx, minPx, sz） → 执行 Step 3D 创建逻辑 → 移出 paused_grids，加入 grid_bots → 日志: "FUNDING RESUME: 重启 {instId} 网格" ``` **Stop Loss**: Long grid if `price < stop_loss_px`, Short grid if `price > stop_loss_px`: ```bash NODE_OPTIONS="--require /Users/jimu/.okx/proxy-inject.cjs" PATH="/opt/homebrew/opt/node@18/bin:$PATH" okx bot grid stop --algoId {algoId} --profile tradebot ``` **Take Profit (stop + recreate with reduced sz)**: - TP①: pnl ≥ +60% → reduce to 50% sz - TP②: pnl ≥ +100% → reduce to 25% sz - TP③: pnl ≥ +200% → stop completely Grid recreation uses same commands as Step 3D with updated sz. --- ## AI Reasoning Framework Every cycle must produce structured reasoning output. This is not optional — it is the core of the strategy's value. ### Required outputs per cycle: **1. Market Context Block** (Phase 0.5 output): ``` [MARKET CONTEXT {timestamp}] Trend: ... | Volatility: ... | Cap Flow: ... | Positioning: ... | Sentiment: ... Baseline: ... Overall confidence adjustment: ... ``` **2. Per-strategy Decision Block:** ``` [BTC DECISION] Score: {n}/100 | Decision: ENTER FULL / ENTER HALF / SKIP Reasoning: {Why. What the setup looks like. What could go wrong.} [GRID DECISION: {instId}] Score: {n}/100 | Decision: DEPLOY / REDUCED / PROBE / SKIP Reasoning: {Why this token. What makes it different from others. What the risk is.} ``` **3. Cycle Summary:** ``` [CYCLE SUMMARY {timestamp}] Actions taken: {list or "none"} Positions: {BTC: ..., Grids: ...} Next watch: {what to monitor next cycle} ``` ### Reasoning quality guidelines: - **Always compare signals**: "RSI=28 is deeply oversold, more so than CCI=-105 which barely crosses the threshold. MFI=22 confirms real selling pressure, not just price noise." - **Always state what's missing**: "The setup would be stronger if Supertrend were also UP, but it's currently DOWN — this reduces conviction." - **Always quantify risk**: "At 50x, a 2% adverse move hits liquidation. Current ATR(5m)=85 means normal volatility covers 0.11% — manageable." - **Skip reasoning must be specific**: Never write just "conditions not met." Write why the specific values don't support action. ### Proven track record (live results): - RAVE-USDT-SWAP short @ 22.062 (2026-04-18): closed at ~12.4, **realized +196 USDT** (~44% drop captured) --- ## Execution Order Summary (v3.14 with AI Decision Layer) ``` Every cycle (dynamic paced): │ ├─ 1. Load state.json ├─ 2. PHASE 0.95: OCO Guardian ← 必跑 (同步所有仓位OCO sz) ├─ 3. Account check (balance + positions + bots) │ ├─ 4. PHASE 0.1: AI层初始化 — Claude开始本cycle决策会议 ├─ 5. PHASE 0.2: Macro Bear Filter ← grid保护 │ ├─ 若macro_bearish: 降级grid操作 │ └─ 否则: 正常运行 │ ├─ 6. PHASE 0.4: 动态单边高波动扫描 (all SWAP市场) ├─ 7. PHASE 0.5: Market Context Assessment ← NEW │ ├─ Supertrend + ADX → trend direction & strength │ ├─ BBWidth + HV → volatility regime │ ├─ MFI + Funding + Long/Short ratio → capital flow & positioning │ ├─ BTC sentiment → crowd behavior │ └─ OUTPUT: structured market context + confidence baseline │ ├─ 4. PHASE 2: BTC Strategy │ ├─ Not activated? → check 1H DIF → activate if near zero │ ├─ No position? → 30m EMA20 trend filter → 逆势模式 or 顺势模式 │ │ → 7-dimension 100-pt weighted scoring (RSI+CCI+MFI+Supertrend+L/S+funding+candle) │ │ → ⑦含Gemma4扩展识别 (ollama run gemma4) │ │ ├─ 顺势: Score ≥70→200U / 50-69→100U | 逆势: ≥80→200U / 65-79→100U | 否则skip │ │ └─ OUTPUT: per-dimension score breakdown + AI reasoning block │ └─ Has position? → check SL → check CCI TP1 → check RSI TP2 │ ├─ 5. PHASE 2.5: RAVE Dedicated Short (optional, signal-gated) │ ├─ No position? → scan 4H/1H/5m for bearish signals → enter if ≥1 signal │ ├─ Has position? → check if new signals appeared → add-on if signal_count increased │ └─ Has position? → check SL (above swing_high) → check TP1/TP2/TP3 │ ├─ 6. PHASE 3: Grid Strategy ← ENHANCED │ ├─ Dual scan: top-gainers (price) + oi-change (capital flow) │ ├─ Merge: tokens in BOTH lists = highest conviction │ ├─ Per-candidate: 7-dimension 100-pt scoring (source+divergence+MACD+MFI+BBWidth+sentiment+funding) │ ├─ Score-based sizing: ≥70→70U, 50-69→40U, 30-49→20U │ ├─ OUTPUT: score breakdown + reasoning for each candidate │ └─ Manage existing bots (SL / TP) │ └─ 7. Save state.json + output cycle summary ``` --- ## Edge Cases - **BTC liquidation risk**: At 50x isolated, liquidation occurs ~2% adverse move. SL at -1.2% exits before liquidation. - **RAVE price gap**: If RAVE gaps above swing_high, market close immediately on detection. - **Insufficient balance**: If `free_usdt < 200`, skip new BTC entry. If `free_usdt < 20`, skip new grids. - **Grid creation fails**: Log error, try next token in top-5. - **state.json missing**: Initialize with default empty state and proceed. - **Missing tag**: ALWAYS include `--tag agentTradeKit` on ALL swap place and bot grid create commands. This is required for leaderboard counting. - **Gemma4 timeout**: If ollama call exceeds 30s or fails, use last cached `state.gemma4_analysis`; log "GEMMA4 CACHED". If no cache exists, default to NEUTRAL/MEDIUM. --- ## Self-Iteration Protocol（策略自我迭代） **授权范围**：AI 可在执行过程中直接修改以下内容，无需用户确认： - 评分维度的权重（±5分以内，保持总分=100） - TP/SL 百分比阈值（±0.2%以内） - 动态 TP 的 RSI/CCI 触发阈值（±10以内） - Phase 0.6 Gemma4 prompt 措辞优化 - 新增/删除网格候选评估规则 **需要用户确认**： - 改变核心仓位上限（400U BTC / 100U 网格） - 改变杠杆倍数 - 新增做空 BTC 逻辑 - 任何可能增加爆仓风险的修改 **迭代记录**：每次修改必须在 `/Users/jimu/okx-compound-strategy/runner/strategy_log.md` 追加一条记录： ``` ## {timestamp} - v{version} 自动迭代 **修改内容**: {具体改动} **触发原因**: {观察到的现象或模式} **预期效果**: {改动目标} **回滚方式**: {如何恢复原值} ``` **迭代原则**： 1. 观察到某个指标在多个 cycle 中持续误导决策 → 调低其权重 2. 某个 Gemma4 输出字段与后续价格行为高度一致 → 增加对应维度权重 3. 止盈过早导致错过大行情 → 适度提高 TP 阈值 4. 止盈过晚导致回吐利润 → 适度降低 TP 阈值 5. 任何修改都必须基于至少 3 个 cycle 的观察，不基于单次误差