القائمة
Use when the user asks to research or investigate a question, claim, or topic in depth. Comprehensive investigation with current information, every claim traceable to a source.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Use when starting project code-improvement work — refactoring, architecture improvement, cleanup of an existing codebase — or when a grilling session should leave docs behind. Runs the interview and records decisions as ADRs and domain terms in CONTEXT.md as they crystallise.
Use when the user requests development work — implementing a feature, debugging an error, designing an approach, or deploying a release. MUST be invoked as the entry point before any implementation, debugging, or design sub-skill.
Use when the user names final QA as the goal — pre-release verification, "上線前驗證 / 最終品質保證".
A relentless interview to sharpen a plan or design.
Grill the user relentlessly about a plan, decision, or idea. Use when the user wants to stress-test their thinking, or uses any 'grill' trigger phrases.
Use when the user asks to check, audit, or review something thoroughly. Builds a check plan, then verifies each item with the investigating skill.
| name | investigating |
| description | Use when the user asks to research or investigate a question, claim, or topic in depth. Comprehensive investigation with current information, every claim traceable to a source. |
| model | sonnet |
Investigate thoroughly. Every claim must be traceable to a source.
search_graph to find symbols, trace_path for call chains, get_code_snippet for source, get_architecture for structure; run index_repository if the project is not indexed) — fall back to grep/read for text, configs, and non-code files; check git history## Findings
### Confirmed
[Evidence-backed conclusions]
### Likely (needs verification)
[Reasonable inferences with stated assumptions]
### Unknown
[What we could not determine and why]
### Sources
[Where each finding came from]
When the question is in code or systems you have not touched, send a scout agent before doing the real investigation.
How:
Explore or general-purpose) with a deliberately loose prompt — "explore X, report what you found, what's confusing, where you'd dig next."
Subagents do NOT receive the session code-discovery reminder, so the prompt MUST include: "Use codebase-memory-mcp tools first (search_graph, trace_path, get_code_snippet, search_code) before grep/file-read."Why this beats jumping straight in: the scout pays the discovery cost in parallel with whatever else you're doing, and the precise prompt that follows hits its target on the first try instead of meandering.
Skip the scout when: the territory is familiar, the question is narrow (one file, one symbol), or the answer is in obvious places (git log, a known doc).