anchor

---

✨ What it does

Anchor puts a personal code reviewer inside your editing loop. Deterministic scripts gather the diff, context, and learnings; the reasoning happens in the Claude Code session you already have open — no API keys, no servers, no per-call cost.

The portable core is the Agent Skill at skills/anchor-review/SKILL.md, which every harness loads. The /anchor command and the push-reminder hook are Claude-Code-native conveniences that some harnesses also pick up.

🚀 Quick start

Install the plugin (Claude Code: via the mjenkins-toolbox marketplace), then in any repo:

/anchor init        # build the codebase map + dependency graph (once)
/anchor review      # review uncommitted changes
/anchor full        # doctor + review + auto-archive
/anchor status      # repo + last-review + git summary

Review targets: (none) = uncommitted · --staged · --since-last · main..feature · pr 123 / pr <url> · @path/to/file.

For per-harness install details and the honesty notes, see Installation.

📚 Documentation

Anchor's full docs live in docs/. Start with the overview, then dig into the workflow and configuration.

Doc	Description
Overview	What Anchor is and how it works
Installation	Per-harness install + updating
Usage	/anchor subcommands, targets, push reminder hook
Review workflow	How a review runs, strictness, and Phase 4 features
Configuration	.anchor/config.yaml keys and per-repo files
Repository layout	What each path in the repo is
Development & release	Build, test, and cut a release
Cross-harness portability	What loads/runs on each harness

Full documentation map: docs/README.md

📄 License

MIT © 2026 Mike Jenkins