# cross-review

> cross-review is an MIT-licensed skill pack from TLE Labs for Claude Code and Codex. The agent you work with writes plans and code; a second model, the Codex CLI today, reviews them round by round until it approves. Six skills: loop-plan, loop-implement, review-plan, review-impl, review-think and review-pr, plus a local dashboard of every review and plan.

- Install with `npx @tlelabs/cross-review` (Node.js 22+ and npm). It installs the skills for Claude Code in `~/.claude/skills/` and for Codex in `~/.agents/skills/`, and registers the `cross-review` MCP server with both.
- The reviewer is the Codex CLI (`codex` in PATH, logged in). Run a skill by name: `/review-impl` in Claude Code, `$review-impl` in Codex.
- Review settings live in `~/.cross-review/config.json` (global) and `<workspace>/.cross-review/config.json`; sessions and plans in `<workspace>/.cross-review/`.
- Each page below is plain Markdown. Treat these docs as the current contract; do not infer unlisted commands, flags or config keys.

## Docs

- [cross-review documentation](https://cross-review.tlelabs.com/docs.md): cross-review adds review skills to Claude Code and Codex. A second model checks your agent's plans, code and reasoning, round by round, until it approves.

## Start here

- [Install cross-review](https://cross-review.tlelabs.com/docs/start/installation.md): Install cross-review with npx, check the runner and the MCP registrations, update it, and move from the codex-review pack.
- [Quickstart](https://cross-review.tlelabs.com/docs/start/quickstart.md): Make a change, ask for a review with review-impl, follow the rounds while they run, read the outcome and open the dashboard.
- [Core concepts](https://cross-review.tlelabs.com/docs/start/concepts.md): The terms cross-review uses: orchestrator, reviewer, session, round, issue, verdict, consensus, stalemate, fix or dispute, and user authority.
- [FAQ](https://cross-review.tlelabs.com/docs/start/faq.md): Answers on cost, models, the reviewer's access, other reviewers, missing Codex login, privacy, the update check and resuming a stopped review.

## Skills

- [loop-plan](https://cross-review.tlelabs.com/docs/skills/loop-plan.md): Write a phased, code-grounded implementation plan for large work, so a fresh session can carry it out without guessing.
- [loop-implement](https://cross-review.tlelabs.com/docs/skills/loop-implement.md): Carry out an approved loop-plan plan phase by phase, with tests first, checkpoint reviews, and one commit per phase.
- [review-plan](https://cross-review.tlelabs.com/docs/skills/review-plan.md): Have a second model review a plan before any code is written, then accept, dispute or escalate each issue until the reviewer approves or you decide.
- [review-impl](https://cross-review.tlelabs.com/docs/skills/review-impl.md): Have a second model check your uncommitted changes, unpushed commits or branch against the plan, then fix the valid issues and rebut the rest until the reviewer approves.
- [review-think](https://cross-review.tlelabs.com/docs/skills/review-think.md): Debate a technical question with a second model, each analyzing it independently first, then converging on a recommendation, a consensus, or an explicit disagreement.
- [review-pr](https://cross-review.tlelabs.com/docs/skills/review-pr.md): Review a pull request you did not write with a second model as a peer, then get one consolidated review with a merge recommendation. Nothing is edited or committed.

## Guides

- [Dashboard](https://cross-review.tlelabs.com/docs/guides/dashboard.md): Start the local cross-review dashboard to follow reviews and plans live, and to change settings, in your browser.
- [Settings](https://cross-review.tlelabs.com/docs/guides/settings.md): Set the default reviewer, model, effort and dashboard switches globally or per workspace, and see which value wins.
- [Plans](https://cross-review.tlelabs.com/docs/guides/plans.md): Where loop-plan writes plans, how loop-implement finds and ticks them, and whether plans are kept in git.
- [Automatic reviews](https://cross-review.tlelabs.com/docs/guides/automatic-reviews.md): Add a block to Claude Code's CLAUDE.md so review-plan and review-impl run after every plan and change, and how to remove it.
- [Troubleshooting](https://cross-review.tlelabs.com/docs/guides/troubleshooting.md): Fix stopped, stalled and timed-out reviews, invalid settings, missing MCP registration, PATH and reviewer install problems.

## Reference

- [CLI reference](https://cross-review.tlelabs.com/docs/reference/cli.md): The cross-review command, its everyday commands, the verbs the skills use, the low-level runner commands, and the exit codes.
- [MCP tools reference](https://cross-review.tlelabs.com/docs/reference/mcp-tools.md): The four cross-review MCP tools, their parameters, results, the next values of review_wait, and the errors they return.
- [Configuration reference](https://cross-review.tlelabs.com/docs/reference/configuration.md): Every key in a cross-review settings file, where it can appear, its default, how values resolve, and the CONFIG_INVALID error.
- [Session files reference](https://cross-review.tlelabs.com/docs/reference/session-files.md): What each review session folder under .cross-review/sessions contains, which file each part of a review writes, and how the folders are ignored by git.

## Optional

- [Complete documentation](https://cross-review.tlelabs.com/llms-full.txt): Every docs page in one Markdown file
- [Source code](https://github.com/lploc94/cross-review): MIT-licensed repository and releases
- [npm package](https://www.npmjs.com/package/@tlelabs/cross-review): Published releases of @tlelabs/cross-review