菜单
Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me".
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
End-to-end PR review pipeline. Detects the PR on the current branch, then runs pr-triage → pr-execute → pr-verify in sequence. Supports resume: if artifacts already exist from a prior run, skips forward to the right phase. No user prompts except the one deferred-ask pause in pr-execute. Precondition: an open PR must exist for the current branch (or provide an explicit PR number).
Snapshot the current state of all work queues and recommend the next skill to run. Reads BUILD_LEARNINGS.md, GitHub Issues (by type: Idea, Bug, Concern), unscheduled issues, PRIORITIES.md, and the git log to assess staleness. Produces a concise status report, then asks the user whether to invoke the recommended skill. Use when resuming work after any interruption, or when unsure what to do next in the development cycle.
Execution phase of the PR review pipeline. Reads .claude/pr-{number}-triage.json, applies all FAIL/IMPROVE fixes inline, remediates CI failures, files GitHub issues for out-of-scope findings, resolves review thread comments, and writes .claude/pr-{number}-execute.json. Use after /pr-triage, or as the second step of /pr-ship.
Pure-discovery phase of the PR review pipeline. Runs all 11 inspection phases (orient through CI status), produces a machine-readable finding list at .claude/pr-{number}-triage.json, and posts a human-readable PR comment. Makes NO code changes. Use as the first step of /pr-ship, or standalone when you want a full finding inventory before deciding whether to execute fixes.
Verification phase of the PR review pipeline. Reads .claude/pr-{number}-execute.json, spot-checks each claimed fix against HEAD (not just the commit diff), validates CI and test suite, and posts a per-finding verdict comment on the PR. Cleans up both artifact files as its final step. Makes NO code changes. Use after /pr-execute, or as the third step of /pr-ship.
DEPRECATED — use /pr-ship (full pipeline) or /pr-execute (fix phase only) instead. Kept as reference for testing logic and comment resolution strategies absorbed into pr-execute/REFERENCE.md; do not invoke directly for new work.
| name | grill-me |
| description | Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me". |
Interview me relentlessly about every aspect of this plan until we reach a shared understanding. Walk down each branch of the design tree, resolving dependencies between decisions one-by-one. For each question, provide your recommended answer.
Ask questions one at a time using the ask_user tool. Never ask questions
in plain text — always use ask_user so the user gets a proper response UI.
Provide a recommended answer (or a choices array with a recommended option
first) for every question.
If a question can be answered by exploring the codebase, explore the codebase instead of asking the user.
