菜单
Use when the user asks to review, debug, rewrite, strengthen, or optimize a prompt; identify ambiguity, missing context, weak constraints, output format issues, evaluation criteria gaps, prompt-injection risk, or hallucination risk; or produce a clearer ready-to-use prompt.
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
| name | prompt-reviewer |
| description | Use when the user asks to review, debug, rewrite, strengthen, or optimize a prompt; identify ambiguity, missing context, weak constraints, output format issues, evaluation criteria gaps, prompt-injection risk, or hallucination risk; or produce a clearer ready-to-use prompt. |
You are an expert Prompt Engineer. Your task is to critically analyze and review the Prompt provided by the user.
Your review should systematically evaluate the prompt across the following dimensions:
Please provide your review in the following structured Markdown format:
[Provide the fully rewritten, optimized prompt here. It should incorporate all the improvements and be ready to use.]
CRITICAL INSTRUCTIONS:
Use when the user asks to design, create, review, improve, package, distribute, measure, or classify AI agent Skills, Claude Code Skills, Codex skills, or reusable agent workflows; convert recurring workflows, runbooks, verification steps, API usage, scaffolding, code review, CI/CD, infrastructure operations, data analysis, or team automation into a maintainable SKILL.md with gotchas, progressive disclosure, resources, evals, rollout guidance, and measurement.
Use when the user asks to write or improve CLI help output, usage text, option/flag documentation, subcommand docs, examples, terminal `--help` text, or a man-page style command reference from raw arguments or an existing CLI.
Use when the user asks to create, revise, or audit a CONTRIBUTING.md/contributor guide, onboarding instructions, local development setup, test/lint workflow, PR process, commit conventions, or open-source contribution guidance for a repository.
Use when the user asks to review, debug, harden, or optimize a Dockerfile or container build for smaller images, faster caching, multi-stage builds, non-root runtime users, dependency installation, .dockerignore, or Docker best practices.
Use when the user asks how to perform, explain, or recover from a Git operation: status/log inspection, branching, staging, commits, amend, reset, revert, rebase, merge conflicts, stash, reflog, submodules, remotes, force-push safety, or rewriting history.
Use when the user asks to triage an issue, bug report, feature request, crash report, reproduction steps, maintainer response, issue priority/severity, missing diagnostic information, or labels for GitHub/GitLab-style issue content.