context-doctor

---

Pick one. They install differently and buy you different things.

① Just the gate — you do not need Claude Code

npx --package @chrono-meta/fh-gate fh-gate          # nothing to install
brew tap chrono-meta/forge-harness && brew install forge-harness   # or this

In GitHub Actions — the same gate as a step, with the verdict kept typed:

- uses: chrono-meta/[email protected]
  with:
    files: ${{ steps.changed.outputs.files }}
  env:
    ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}

Listed on the GitHub Actions Marketplace.

The step exposes verdict (PASS · PENDING · BLOCKED · ESCALATE · HARNESS_ERROR · ARG_ERROR · DRY_RUN · UNKNOWN) and reviewed. reviewed: false is not a pass — a backend that never answered, a dry run, or an exit code this wrapper does not know all land there, and all of them fail the step by default. That default is the point: a check that did not run must never read green. Change it with fail-on: if you want a softer policy, and know what you are trading.

What you get

• A change is judged before it merges, and the verdict names what the change lost — not that something is "off". The GIF above is that verdict on a real diff.
• The verdict is a typed value, not text you grep: PASS · PENDING · BLOCKED · ESCALATE.
• Runs anywhere a shell runs — CI, a pre-commit hook, a different coding agent. Claude Code optional.
• It reviews your own code, not just an agent's. Point it at a diff and it names the weakness — a verdict that quietly degrades toward PASS, a reference that does not exist, a secret, a claim with no grounds — so you fix it and re-run before merge. Where each FH engine applies (harness building · skill/agent authoring · code review · irreversible-surface gates · context continuity), and what changes by model tier and effort level: docs/USE_CASES.md · docs/model_tier_expectations.md. How the gates line up with ISO/IEC AI-testing and AI-quality standards (42119 · 29119-11 · 25059 · 42001), as a self-assessment with evidence pointers: docs/STANDARDS_ALIGNMENT.md.

② The whole harness — inside Claude Code

claude plugin marketplace add https://github.com/chrono-meta/forge-harness.git
claude plugin install -s user fh-meta@forge-harness
claude plugin install -s user fh-qp@forge-harness      # optional: QP (Quality Platform) — plan→run→regress a web/desktop app through the session's Playwright / computer-use MCP
git clone https://github.com/chrono-meta/forge-harness.git ~/projects/forge-harness
cd ~/projects/forge-harness && claude        # then type a greeting: hi · 안녕 · こんにちは · 你好

What you get, on top of ①

• You stop having to pick the check. The harness reads what you are about to do — publish, delete, rewrite history, open a PR — and names the gate for that moment. ① is one command you remember; ② is the layer that remembers for you.
• 46 skills · 14 agents you can call in plain language: diagnose a project, accelerate one, wire a new one up.
• tracks/ keeps what each session learned, so session 2 starts where session 1 stopped. This is the part that compounds — and the part you cannot judge on day one.
• Ask for the same thing three times and it stops answering: it builds you the harness that answers.

🟥 One thing ② does not give you. FH also carries a 4-axis pre-commit hook, and it is not for your repos: it hard-codes hub paths and hub markers, so installing it into your project blocks your commits instead of helping. The installer treats it as opt-in and tells you to skip it unless you develop FH itself. For your own repos, ① is the gate — wire it into CI or your own pre-commit.

Not sure? Start with ①. It costs one command and nothing to uninstall, and ② is a superset — nothing you learn in ① is thrown away.

What neither door is

It does not replace the review that happens after. It moves the question earlier, so that what reaches a human reviewer is smaller — not so that a human stops reviewing. The bottleneck it targets is the gap between how fast things get generated and how fast a person can check them; it closes that gap from the front, by cutting what has to travel to the back.

What a diff cannot show stays a person's job. Anything that only surfaces when the thing actually runs — on a real screen, aga