fix(agent): include framework category in skill rendering + Docker strategies missing (#406)

* fix(agent): include framework category in skill instructions rendering

`get_skill_instructions()` only iterated over ["trend", "pattern",
"reversal"] categories, silently dropping all strategies with
`category: framework` (box_oscillation, chan_theory, wave_theory,
emotion_cycle). Add "framework" to the category map and use a
dynamic fallback so future custom categories are never lost.

Also:
- Document `AGENT_SKILLS=all` shorthand in .env.example
- Add `AGENT_SKILLS` config to README
- Add CHANGELOG entry
- Update strategies/README.md and Skill docstring with framework category

Fixes #403

* fix(docker): include strategies directory in Docker image and compose

Dockerfile was missing `COPY strategies/ ./strategies/`, and
docker-compose.yml did not mount the strategies directory, causing
all 11 built-in strategies to fail loading in containerized deployments.

---------

Co-authored-by: ma_k <ma_k@ctrip.com>

.env.example

@@ -127,7 +127,8 @@ BRAVE_API_KEYS=your_brave_key_here
 # 推荐：趋势 + 金叉 + 缩量回踩（兼顾多种市场形态）
 # AGENT_SKILLS=bull_trend,ma_golden_cross,shrink_pullback
 #
-# 完整启用所有内置策略：
+# 完整启用所有内置策略（两种写法等效）：
+# AGENT_SKILLS=all
 # AGENT_SKILLS=bull_trend,ma_golden_cross,volume_breakout,shrink_pullback,bottom_volume,dragon_head,one_yang_three_yin,box_oscillation,chan_theory,wave_theory,emotion_cycle
 #
 AGENT_SKILLS=bull_trend,ma_golden_cross,volume_breakout,shrink_pullback

README.md

@@ -138,6 +138,7 @@
 | `NEWS_MAX_AGE_DAYS` | 新闻最大时效（天），默认 3，避免使用过时信息 | 可选 |
 | `BIAS_THRESHOLD` | 乖离率阈值（%），默认 5.0，超过提示不追高；强势趋势股自动放宽 | 可选 |
 | `AGENT_MODE` | 开启 Agent 策略问股模式（`true`/`false`，默认 false） | 可选 |
+| `AGENT_SKILLS` | 激活的策略（逗号分隔），`all` 启用全部 11 个；不配置时默认 4 个，详见 `.env.example` | 可选 |
 | `AGENT_MAX_STEPS` | Agent 最大推理步数（默认 10） | 可选 |
 | `AGENT_STRATEGY_DIR` | 自定义策略目录（默认内置 `strategies/`） | 可选 |
 | `TRADING_DAY_CHECK_ENABLED` | 交易日检查（默认 `true`）：非交易日跳过执行；设为 `false` 或使用 `--force-run` 强制执行 | 可选 |

docker/Dockerfile

@@ -47,6 +47,7 @@ COPY data_provider/ ./data_provider/
 COPY bot/ ./bot/
 COPY patch/ ./patch/
 COPY src/ ./src/
+COPY strategies/ ./strategies/
 COPY --from=web-builder /app/static ./static/
 
 # 创建数据目录

docker/docker-compose.yml

@@ -24,6 +24,7 @@ x-common: &common
     - ../logs:/app/logs
     - ../reports:/app/reports
     - ../.env:/app/.env
+    - ../strategies:/app/strategies:ro
     # 如需覆盖前端静态资源，可挂载本地 static 目录
     # - ../static:/app/static:ro
 

docs/CHANGELOG.md

@@ -35,6 +35,11 @@
   - 扩展 `analysis_tools` 与 `data_tools`，优化策略问股的工具调用链路与分析覆盖
 
 ### 修复（#patch）
+- 🐛 **Agent 策略渲染遗漏 framework 分类**（Issue #403）
+  - 根因：`get_skill_instructions()` 仅遍历 `trend/pattern/reversal` 三个分类，`category: framework` 的 4 个策略（箱体震荡、缠论、波浪理论、情绪周期）被静默丢弃
+  - 修复：补充 `framework` 分类，并增加动态回退机制，确保未来自定义分类不会遗漏
+  - 文档：`.env.example` 补充 `AGENT_SKILLS=all` 写法，`README.md` 配置表新增 `AGENT_SKILLS`
+  - Docker：Dockerfile 补充 `COPY strategies/`，docker-compose.yml 挂载 `strategies/` 目录（此前容器内策略目录缺失，导致所有策略均无法加载）
 - 🐛 **支持 DeepSeek 思考模式**（Issue #379）
   - 根因：Agent 模式（tool calls）下使用 DeepSeek 思考模式时，未在 assistant 消息中回传 `reasoning_content`，导致 API 返回 400
   - 修复：`llm_adapter._call_openai` 解析并透传 `reasoning_content`；`executor` 在 assistant_msg 中写入该字段

src/agent/skills/base.py

@@ -35,7 +35,7 @@ class Skill:
         display_name: Human-readable name (e.g., "龙头策略").
         description: Brief description of when to apply this strategy.
         instructions: Detailed natural language instructions injected into the system prompt.
-        category: Strategy category — "trend" (趋势), "pattern" (形态), "reversal" (反转).
+        category: Strategy category — "trend" (趋势), "pattern" (形态), "reversal" (反转), "framework" (框架).
         core_rules: List of core trading rule numbers this strategy relates to (1-7).
         required_tools: List of tool names this strategy depends on.
         enabled: Whether this strategy is currently active.
@@ -254,7 +254,7 @@ class SkillManager:
             return ""
 
         # Group by category
-        categories = {"trend": "趋势", "pattern": "形态", "reversal": "反转"}
+        categories = {"trend": "趋势", "pattern": "形态", "reversal": "反转", "framework": "框架"}
         grouped: Dict[str, List[Skill]] = {}
         for skill in active:
             cat = skill.category or "trend"
@@ -262,7 +262,9 @@ class SkillManager:
 
         parts = []
         idx = 1
-        for cat_key in ["trend", "pattern", "reversal"]:
+        # Render known categories in fixed order, then any remaining custom categories
+        ordered_keys = ["trend", "pattern", "reversal", "framework"]
+        for cat_key in ordered_keys + [k for k in grouped if k not in ordered_keys]:
             skills_in_cat = grouped.get(cat_key, [])
             if not skills_in_cat:
                 continue

strategies/README.md

@@ -26,7 +26,7 @@ name: my_strategy
 display_name: 我的策略
 description: 简短描述策略适用的市场场景
 
-# 策略分类：trend（趋势）、pattern（形态）、reversal（反转）
+# 策略分类：trend（趋势）、pattern（形态）、reversal（反转）、framework（框架）
 category: trend
 
 # 关联的核心交易理念编号（1-7），可选