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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/yamadashy/skills --skill claude-review-loop命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Consult OpenAI Codex as an advisor via `codex exec` — a cross-model second opinion on a hard design decision, a stuck debugging session, or a close review call. A consultation, not an implementation hand-off. Triggers on: 'consult codex', 'ask codex for advice', 'get codex's opinion', 'second opinion from codex', or when a judgment call is too close to make alone.
Consult Fable (claude-fable-5, the tier above Opus) as an advisor — a second opinion on a hard design decision, a stuck debugging session, or a close review call. From Claude Code, spawn a subagent on model fable; from any other harness, call the Claude Code CLI. A consultation, not an implementation hand-off. Triggers on: 'consult fable', 'ask fable', 'get fable's opinion', 'second opinion from fable', or when a judgment call is too close to make alone.
Shape what a message says around the recipient's state rather than the sender's. Corrects the measured sender biases first (underestimating that people say yes, overestimating how well you were understood, arguing from your own values), then channel and friction, then barriers to comprehension, and wording last. Load-bearing claims carry an evidence grade, and popular techniques that the evidence does not support — or that fail an honesty test — are named and excluded. Use when drafting or revising a spoken or written request, proposal, explanation, bad news, apology, or a reply in a disagreement — and on "make this more persuasive", "how should I phrase this", "they'll say no", "this didn't land". Changes substance and stance — not layout, not scannability, not prose rhythm.
正在显示 SKILL.md
基于 SOC 职业分类
| name | claude-review-loop |
| description | Iterative Claude review-and-fix loop |
Repeat the following cycle on the current branch's changes against main (max 3 iterations):
claude-exec skill for flag details):
claude -p --no-session-persistence --permission-mode default \
--disallowedTools Bash,Edit,Write \
-- "Review the current branch's diff against main (git diff main...HEAD). Report only real defects — bugs, security issues, broken logic, missing edge cases. For each finding give file:line, severity, and a one-line fix. Skip style nitpicks and scope creep."
npm run lint and npm run test. Fix any regressions and repeat this step until all checks pass before continuing.Stop when no "Fix" items remain or 3 iterations are reached. Print a summary of what was fixed and what was skipped.