用于 Quantia 仓库的"验证优先"诊断与修复工作流——当用户报告 bug、异常、生产日志报错、或要求"检查/审查当前修改是否引入新问题"时使用。强制：先证实问题真实存在并定位根因 → 评估多方案择优 → 修复 → 回归自检 → 本地/接口验证，最后才声明完成。覆盖生产日志（阿里云 115.29.213.22）根因定位、phase/文档一致性核对。纯新功能开发、纯查询/解释、用户已明确"直接改"时可跳过严格证实步骤。
Use when debugging AI agent issues: empty reports, tool call failures, LLM hallucinated columns, SSE streaming errors, max_rounds exhaustion, or provider connectivity. Covers diagnostic steps, common root causes, and fix patterns for the stock_analyst/market_summarizer/strategy_coder agents.
Use when adding, editing, debugging, or reviewing custom composite indicators (Phase 9 自定义综合指标) under quantia/core/composite/, including normalizers (n_lin/n_wr/n_rank/n_supertrend/n_pctb/n_cci), hard-rule AST sandbox expressions, risk simulator stop/target/max-hold + fundamentals exit, dynamic universe filters, and the cn_stock_custom_indicator handler. Use when writing builtins or wiring the customIndicatorHandler / refresh_composite_universe cron.
Use after editing Quantia backend Python (quantia/web/**, quantia/core/**, quantia/job/**) or frontend (quantia/fontWeb/src/**) to restart the long-running Tornado web service, sync template changes, and copy the Vite dist into quantia/web/static for production. Use when changes appear to take no effect in the browser or when frontend strategy templates show stale code.