Gatekeep

AI-assisted PR review for your own repos: parallel multi-model review across providers, fail-closed PR risk tiers, agreed-only inline findings, blocking security check.

中文说明： README.zh-CN.md

Install (remote uses:)

# .github/workflows/gatekeep.yml  ← must live under .github/workflows/ on Gitea/GitHub
- uses: actions/checkout@v4
- uses: https://git.potatolab.uk/layue13/gatekeep@main
  # placeholder: uses: https://git.example.com/OWNER/gatekeep@v1
  env:
    # Names must match providers.yaml `key` — FREE-FORM (no required gatekeep_ prefix)
    OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
  with:
    github-token: ${{ secrets.GITHUB_TOKEN }}

Dogfood (this repo)

layue13/gatekeep reviews its own PRs via .github/workflows/gatekeep-pr.yml with uses: ./ and secrets.OPENROUTER_API_KEY. Open a PR against main to see a review (Gitea Actions must have a runner — same pool as other layue13 repos).

Because dogfood uses uses: ./, rebase/merge main into the PR branch after Action fixes land, or the PR still runs the old Action. Re-run via new push or Actions → Gatekeep → Run workflow. Free-model 429s are retried; extras in models: after the first 3 are fallbacks; risk is force-high only if zero models succeed.

Allowlists / Gitea-on-github.com limitations: docs/remote-uses.md.

Providers YAML

language: zh-CN               # report language: zh-CN | en | … (prompts + published comments)

providers:
  openrouter:                 # only include providers you configure
    key: OPENROUTER_API_KEY   # free-form secret/env name — pick any
    protocol: openai          # openai | openai-completions | openai-responses | anthropic
    # base_url: optional (defaults exist for openrouter / anthropic / openai)
    models:
      - cohere/north-mini-code:free
      - nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free
      - google/gemma-4-31b-it:free

Field	Meaning
language	Report language for model prose + summary/inline comments (zh-CN, en, …). Alias: report_language.
key	Free-form name of the GitHub secret / env var holding the API key. No required gatekeep_ prefix.
protocol	openai | openai-completions | openai-responses | anthropic
models	Names to use (validated against the provider’s model list API at runtime)
base_url	Optional; known provider ids get defaults

Report language

• This repo’s defaults: language: zh-CN in defaults/providers.yaml.
• Override without editing YAML: Action input language: en (or zh-CN).
• Resolution order: Action input language → YAML language / report_language → en.
• JSON field names and enums stay English; human-facing prose follows the configured language.

Mix providers in one review by listing several entries. Secrets are read from the workflow step env using each entry’s key name.

Example suggestion only (still free-form): key: gatekeep_openrouter is fine if you named your secret that — not required.

Finding format

Severity (Critical→Low) + where (path:line) + why it matters. Inline only when ≥2 models agree.

PR risk tiers

Tier	Humans	Auto-merge
low	0 after Gatekeep + security green	If kill switch on
medium	1	No
high	2	No

Optional consumer config

config-dir: .github/gatekeep with your providers.yaml. Omit to use Action defaults/providers.yaml.

Layout

action.yml
scripts/            # resolve-providers.py, call-model.sh, …
defaults/providers.yaml
config/providers.yaml   # example override
workflows/  templates/  evals/  incidents/  docs/