メニュー
Use when the user wants a plan or design stress-tested before implementation, or asks to be grilled, interviewed, or interrogated about a decision tree.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Disciplined root-cause investigation for hard bugs and performance regressions, with findings pinned to the case board. Use when user reports a bug, says something is broken/throwing/failing, describes a performance regression, or asks to diagnose/debug/investigate something.
Use when writing a commit message — "write a commit", "commit message", "generate commit", "/commit", or invokes /faceless-commit, or when staging changes. Legacy alias: "/caveman-commit".
Use when reviewing code or a diff — "review this PR", "code review", "review the diff", "/review", or invokes /faceless-review. Legacy alias: "/caveman-review".
Use when the user wants terse, compressed, low-token responses — "faceless mode", "talk faceless", "be brief", "less tokens", "compress", or invokes /faceless. Also when token efficiency is explicitly requested. Legacy aliases: "caveman mode", "use caveman".
Use when turning raw material — research, notes, interview answers, fragments, or investigation findings — into a sourced written report or article, or when shaping, drafting, restructuring, or tightening such a draft. Triggers: "write up the findings", "investigative report", "turn these notes into an article", "draft the report", "shape this into an article", "edit the article", "help me write this up".
Sets up an `## Agent skills` block in AGENTS.md/CLAUDE.md and `docs/agents/` so the engineering skills know this repo's issue tracker (GitHub or local markdown), triage label vocabulary, and domain doc layout. Run before first use of `to-issues`, `to-prd`, `triage`, `investigate`, `tdd`, `improve-codebase-architecture`, or `zoom-out` -- or if those skills appear to be missing context about the issue tracker, triage labels, or domain docs.
| name | interrogate-spec |
| version | 1.0 |
| category | planning |
| execution_speed | fast |
| token_efficiency | high |
| triggers | ["grill me","stress-test plan","stress test design","interview relentlessly"] |
| cache_key | interrogate-spec-1.0 |
| dependencies | [] |
| disable-model-invocation | true |
| description | Use when the user wants a plan or design stress-tested before implementation, or asks to be grilled, interviewed, or interrogated about a decision tree. |
Replaces grill-me. Reads ${CLAUDE_PLUGIN_ROOT}/QUESTION.md for voice
before starting -- see that file for tone; everything below is mechanics.
Default behavior is the loop above -- one question, one recommendation, move on. When an answer is vague, hand-waved, or contradicts a decision already resolved earlier in the same session, chain exactly one follow-up "why" before accepting it and moving to the next branch.
Never chain more than one follow-up. Never escalate on an answer that's already specific and consistent -- that's interrogation theater, not precision.
If QUESTION.md is missing or unreadable, run the loop above anyway.
Function over costume -- this skill works with or without the voice doc.
When a branch resolves -- the answer is specific, internally consistent, and not contradicted by codebase evidence or prior decisions -- update provenance to confidence: corroborated, method: quoted. This is the spec-decision-locked emit; it fires at most once per branch.
