一键导入
investigation
Fires when investigating root causes or bugs. Performs exhaustive code-path enumeration and counter-evidence checking.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Fires when investigating root causes or bugs. Performs exhaustive code-path enumeration and counter-evidence checking.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Fires when deciding whether a report or explanatory document needs figures, when designing a figure, or when writing the figures declaration in frontmatter. Designs cognitive-load-reducing figures via the trigger table, quantitative limits, and templates
Fires when generating or verifying deterministic HTML from report MDs. Runs render / verify / lint / selftest / check via the bundled md2html.cjs
Fires when producing a report or completion notice. Generates a report that includes the interlock verification section.
Discipline for preventing the stop bug (malformed tool call -> self-priming). Fires when the stopbug-observe hook emits a trip-wire (migration advice) or the time-based timer (calibration prompt), or as a minimal focus-check right before heavy processing.
報告書・説明文書で図を使うか判断する際、図を設計する際、frontmatter の figures 宣言を書く際に発火する。図トリガー表・定量上限・雛形で「認知負荷を減らす図」を設計する
報告書 MD から決定論 HTML を生成・検証する際に発火する。md2html.cjs(同梱)で render / verify / lint / selftest / check を実行する
| name | investigation |
| description | Fires when investigating root causes or bugs. Performs exhaustive code-path enumeration and counter-evidence checking. |
This skill is the procedural discipline of investigation (counter-evidence checks, exhaustive code-path enumeration, the no-guessing rule). It is NOT a symptom-indexed catalog of failure modes.
When you observe a bug, defect, or unexpected behavior, fire the debug-toolkit skill in parallel and look up the menu of known failure modes (FMs) by symptom.
But the debug-toolkit menu is also not exhaustive. When you encounter a symptom that is not on the menu, do not force-fit it to the closest FM (no heuristic over-fit). Build your own SST (Special Service Tool) when needed, and propose adding the technique to debug-toolkit in your completion report if it is generalizable.
Fires whenever you are investigating a root cause, debugging, or isolating a problem.
Actively search for evidence that disproves your own conclusion.
## Counter-evidence check
- [ ] Is there another code path with the same functionality?
- [ ] Did you back up the conclusion with runtime logs?
- [ ] If this conclusion is wrong, what is the real cause? (at least one alternative hypothesis)
- [ ] Are there past cases where the initial conclusion was wrong?
=== Investigation Summary ===
- Cause:
- Scope of impact:
- Fix strategy: