Claude review

Skill Kenmege/codex-claude-companion/skills/claude-review

Use when the user asks Codex to orchestrate Claude for a coding job in another terminal, or to review code, run adversarial or ship/no-ship review, audit security risks, or inspect a large diff/folder.From its SKILL.md

Install

npx -y skills add Kenmege/codex-claude-companion --skill claude-review

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.3 KB, as published. Nobody here has run it

Claude Review

Use the local helper to route the requested Claude lane. New coding work uses a durable tmux-owned bridge worker with recorded collaboration, recovery, delivery, and independent verification. The older workspace lane remains available for compatibility. Review lanes snapshot the target, use read-only agentic mode, validate structured output, and render the result.

Default command

codex-claude review -- "$ARGUMENTS"

Route by intent

Coding or implementation: codex-claude delegate --verify-command '<JSON argv>' -- "$ARGUMENTS"; keep the active Codex task responsive and supervise the returned ccb_... job with wait, status, logs, send, and recover.
Legacy interactive workspace: codex-claude workspace -- "$ARGUMENTS"; use workspace-status, workspace-logs, and workspace-stop.
Everyday diff review: codex-claude review -- "$ARGUMENTS"
Release gate / ship-no-ship: codex-claude review --preset ship -- "$ARGUMENTS"
Security review: codex-claude review --preset security -- "$ARGUMENTS"
Research or evidence-heavy folder: codex-claude folder <path> --preset research --long-context -- "$ARGUMENTS"
Deep multi-agent review: codex-claude review --preset deep --background -- "$ARGUMENTS"

Operating rules

Keep review workflows read-only. Use a writable bridge or workspace lane only when the user asks for coding or implementation.
Require an origin-supplied repository-native verification command for every delegation. Do not accept a worker-chosen command as independent verification.
Treat trusted-autonomous as explicit cooperative same-UID host trust, not a sandbox. Never claim sandbox-autonomous is available.
The bridge may launch ephemeral Codex for independent verification. Do not hardcode a GPT model or claim the worker verified itself independently.
Return the helper output verbatim.
If the helper exits non-zero, report the exact failure and stop.
For first-run problems, ask the user to run codex-claude enable and codex-claude doctor.