AgentSkillsCN

code-review

同时启用 claude-reviewer（Opus）与 codex-reviewer（Codex CLI）代理，开展并行代码审查。仅在用户明确要求进行代码审查时才予以调用。

SKILL.md
--- frontmatter
name: code-review
description: "Launch parallel code reviews using both claude-reviewer (Opus) and codex-reviewer (Codex CLI) agents. Only invoke when the user explicitly requests code review."
disable-model-invocation: true

Launch both reviewer agents in parallel for comprehensive code review.

Usage

code
/code-review [PR#] [uncommitted]
  • •No args: Review committed changes against upstream (origin/main or origin/master)
  • •PR# or URL: Review a specific pull request
  • •uncommitted: Include uncommitted/staged changes

Execution

Launch both agents in parallel using the Task tool:

  1. •

    claude-reviewer (subagent_type: claude-reviewer)

    • •Uses Opus model
    • •Reads code directly with Glob/Grep/Read tools
  2. •

    codex-reviewer (subagent_type: codex-reviewer)

    • •Uses Codex CLI with gpt-5.2-codex
    • •Extended reasoning (xhigh)

Pass $ARGUMENTS to both agents verbatim.

Output

Present results under clear headers:

code
## Claude Review (Opus)
[claude-reviewer output]

## Codex Review
[codex-reviewer output]

If both find no issues, state the code looks solid.