菜单
Orchestrator. Triages a PR, local diff, patch file, or staged/branch changes into one of four review types (content, code-lite, code-deep, infra) and dispatches to the matching child skill. Use this when you don't already know which depth or type of review you want. For a known type, invoke the child skill directly.
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。
Computes the optimal Claude usage-window ping schedule from work hours and creates or reconciles the scheduled cloud routines that pin the window boundaries.
Reviews changes to prose files (markdown, text, blog posts, docs). Focuses on voice, grammar, factual claims, dead links, and structure. Loads the project's voice or style guide when one exists. Skips linter and Actions hash verification.
Full code review for large or security-sensitive changes. Gathers requirements first, loads all relevant language references, runs the linter where applicable, applies the full review checklist with security as the top priority.
Reviews infrastructure changes: GitHub Actions workflows, Terraform, Dockerfiles, docker-compose, IaC. Focuses on supply-chain pinning, secret handling, least-privilege permissions, action versions, and CI failure modes. Always verifies action hash pins when present.
Abbreviated code review for small or low-stakes code changes. Skips requirements gathering. Skips the linter in PR mode (CI runs it). Focuses on correctness against obvious intent, readability, and pattern consistency. For large or security-sensitive diffs, use code-review-deep instead.
Produces a one-shot devil's-advocate critique of a plan, design, or decision: steel-manned strengths first, then severity-tagged findings each with a concrete alternative, then a verdict.
| name | code-review |
| description | Orchestrator. Triages a PR, local diff, patch file, or staged/branch changes into one of four review types (content, code-lite, code-deep, infra) and dispatches to the matching child skill. Use this when you don't already know which depth or type of review you want. For a known type, invoke the child skill directly. |
| when_to_use | Use when the user asks for a code review in any form: /code-review, 'code review this', 'CR this', or similar, and the type is not pre-specified. Follow all steps in order; do not shortcut based on this description. |
| argument-hint | [PR URL | path to diff file] [--type=<content|code-lite|code-deep|infra>] [--lint] |
| effort | low |
| allowed-tools | Bash(git *) Bash(gh *) Bash(bash ~/.claude/skills/code-review/scripts/code-review-state.sh *) Read Grep AskUserQuestion Skill |
Read ~/.claude/skills/code-review/references/review-shared.md and follow its source detection and diff fetching sections. By the end of this step you have:
pr / uncommitted / branch / both / fileFrom the diff and the changed file list, compute:
| Signal | How |
|---|---|
lines_added | count of ^+ lines (excluding +++ headers), excluding lockfiles and build-output paths |
files_changed | unique +++ b/ paths, excluding lockfiles |
all_docs | every changed path matches \.(md|mdx|txt|rst)$ or lives under docs/, content/, posts/ |
has_infra | any path under .github/, ends in .tf, matches Dockerfile, or matches docker-compose* |
has_security_paths | any path matches (auth|crypto|password|token|secret|\.env|sql|migration) |
Apply recommendation logic (first match wins):
has_infra true: recommend code-review-infraall_docs true: recommend code-review-contenthas_security_paths true OR lines_added >= 500: recommend code-review-deepPrint this classification line before proceeding:
Triage: <lines_added> lines / <files_changed> files / <comma-separated signals that fired, or "no special signals">. Recommended: ().
Resolve state location. Run:
bash ~/.claude/skills/code-review/scripts/code-review-state.sh resolve
If output is non-empty, that's the state_file path. If empty, persistence is skipped for this run (no project context, or .claude/ missing).
Compute source_id:
pr:<URL>file:<absolute path>git:<mode>:<branch>, where <mode> is one of branch, uncommitted, both, and <branch> is the output of git branch --show-current (use detached if empty)Pick the type, in this order:
--type=<valid> was passed. Use that type. If the value is not one of content, code-lite, code-deep, infra, ignore it and continue to step 2. If state_file is set, run:
bash ~/.claude/skills/code-review/scripts/code-review-state.sh set <state_file> <source_id> <type>
Skip the prompt.
Stored type exists. If state_file is set, run:
bash ~/.claude/skills/code-review/scripts/code-review-state.sh get <state_file> <source_id>
If the output is non-empty, use that type and print:
Using previously chosen review type: . Pass
--type=<other>to override.
Skip the prompt.
Prompt the user. Call AskUserQuestion:
(Recommended) appended to the label, then the other three in this fixed order: code-deep, code-lite, infra, content (skipping whichever is recommended).After the user picks, if state_file is set, run the set command from step 1 to persist the choice.
The chosen label maps to a child skill: content -> code-review-content, code-lite -> code-review-lite, code-deep -> code-review-deep, infra -> code-review-infra.
Invoke the chosen child skill via the Skill tool. Build the args string from the resolved source so the child does not re-prompt:
--source=pr --pr=<URL>--source=file --file=<path>--source=<mode>--lint, append --lintThe child performs the actual review and produces all output. Do not produce any review content yourself: your job ends after dispatching.