菜单
Review code changes for quality, security, and correctness. Use when the user says "review this PR", "review these changes", "check my code", "look at what I changed", or after implementing a feature. Produces a severity-organized report.
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
Robust, line-based file editing. Replace a range of lines in a file with new content by line number instead of exact string matching. Use when the user says 'robust-edit', '/robust-edit', 'replace lines', 'edit lines N to M', or when character-level matching with the Edit tool keeps failing on multi-line changes or uncertain whitespace.
Diagnose and fix context window issues. Use when Claude seems confused, forgetful, slow, or the user says "context is full", "you forgot", "you're repeating yourself", "start fresh", or "why are you ignoring my instructions".
Structured git workflow for commits, branches, and PRs. Use when the user says "commit", "create a PR", "new branch", "prepare for review", "ship this", or "push my changes". Safe, conventional git operations with checkpoints.
Deep interview to reach shared understanding before building. Use when starting a complex feature, when requirements are unclear, or when the user says "grill me", "interview me", "ask me questions", "let's figure this out", or "grill me lightly" for a quick version.
Approval gate for dangerous operations. Use internally before executing destructive commands, file deletions, git pushes, or external API calls from the web chat. Do NOT invoke directly.
Create, list, and manage scheduled tasks and automations. Use when the user asks to schedule something, set a reminder, create a recurring task, or manage existing automations. Trigger phrases: 'remind me', 'every hour', 'schedule', 'at 9am', 'recurring', 'automate', 'cron'.
| name | code-review |
| description | Review code changes for quality, security, and correctness. Use when the user says "review this PR", "review these changes", "check my code", "look at what I changed", or after implementing a feature. Produces a severity-organized report. |
| argument-hint | [file path, branch, or PR number] |
| context | fork |
| agent | Explore |
| allowed-tools | Read, Grep, Glob, Bash(git *) |
Produce a severity-organized review report covering security, correctness, and quality for code changes.
Identify issues in code changes, organized by severity (Critical > Warnings > Suggestions), with specific file:line references and fix suggestions for every finding. Done when the report is presented to the user.
Before reviewing, read CLAUDE.md and any linter/formatter config files (.eslintrc, pyproject.toml, .prettierrc, etc.) to understand project conventions. Do not flag style issues already handled by configured tooling.
$ARGUMENTS specifies a file, branch, or PR: use that scope$ARGUMENTS is empty: run git diff main...HEAD. If that returns nothing, ask: "What should I review — a branch, specific files, or uncommitted changes?"CHECKPOINT: Present the identified files and diff scope. Confirm this is the correct review target before proceeding.
Critical (must fix before merge):
Warnings (should fix):
Suggestions (nice to have):
For each finding:
path/to/file.ts:42Organize findings by severity: Critical first, then Warnings, then Suggestions.
Report is presented inline in the conversation, organized by severity. Offer to save to .claude/memory/review-YYYY-MM-DD.md if the user wants a record.