Agent Friendly Code

Agent Friendly Action — score every PR

A GitHub Action that comments the agent-friendliness delta on every pull request — “this PR drops your Claude Code score by 4.1 points because it removed CI config”. Runs entirely inside your CI, no third-party server, opt-in via a single secret.

Catch score regressions on every PR

Drop the workflow below into your repo and the action will post (or edit) a single comment on each PR with the score delta and the per-signal changes — runs entirely inside your CI, no third-party server in the loop.

# .github/workflows/agent-friendly.yml
name: Agent-friendly score diff

on:
  pull_request:
    branches: [main]

permissions:
  contents: read
  pull-requests: write

jobs:
  score-diff:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0
      - uses: hsnice16/agent-friendly-action@v0
        with:
          agents-badge-token: ${{ secrets.AGENTS_BADGE_TOKEN }}

Set AGENTS_BADGE_TOKENin the repo's secrets to enable the comment; leave it unset and the action exits silently.

Set the AGENTS_BADGE_TOKEN secret

The action only fires when AGENTS_BADGE_TOKEN is present — it's an opt-in marker, not a credential. Add it once and forget it:

  1. In your repo, open Settings → Secrets and variables → Actions.
  2. Click New repository secret.
  3. Name: AGENTS_BADGE_TOKEN. Value: any non-empty string — enabled works. The string itself isn't checked.
  4. Click Add secret. The action fires on the next PR.

Authoritative reference: docs.github.com — encrypted secrets .

How it works

  1. actions/checkout with fetch-depth: 0 gives the action the PR head and access to base history.
  2. The action fetches the base ref locally and checks it out into a worktree — no network clone.
  3. Scores both trees with the bundled scorer and computes the overall + per-signal + per-model delta.
  4. Posts (or edits) a single PR comment marked with <!-- agent-friendly-action -->.

Self-contained by design

The scorer and weights are bundled into dist/ via @vercel/ncc. If Agent Friendly Code goes offline tomorrow, the action keeps producing PR comments unchanged. Runtime weight refresh from a future /api/weights endpoint is deferred to v1.0.0 once benchmark-derived weights ship — until then both the head and the base are scored with the same bundled weights, so any drift cancels in the diff.

FAQ

  • What does the action do on a PR?

    It runs inside your CI (no third-party server in the loop), checks out the PR head and the base ref, scores both trees with the bundled scorer, and posts a single PR comment with the overall delta plus per-signal changes. On the next push it edits that same comment instead of creating a new one.

  • Why is AGENTS_BADGE_TOKEN required?

    It's an opt-in marker, not an API credential. Set it (any non-empty string) to enable the comment; leave it unset and the action exits silently. This lets template / starter repos ship the workflow without it firing for forks or unenrolled installs.

  • Does it talk to your web app?

    No. The scorer and weights are bundled into the action's dist via @vercel/ncc. If this dashboard goes offline, the action keeps producing PR comments unchanged. A future version will optionally fetch fresh weights from /api/weights once benchmark-derived weights ship (1.0.0).

  • Does it work on private repos?

    Yes — it runs in your CI under your existing GITHUB_TOKEN. The action never sends repo contents anywhere; the scoring is local to the runner.

  • Where is the source?

    github.com/hsnice16/agent-friendly-action. MIT-licensed, semver-tagged. Pin @v0 to track the latest 0.x release; pin @v0.1.0 to opt out of automatic minor / patch updates.

Source

github.com/hsnice16/agent-friendly-action — MIT-licensed, semver-tagged. Listed on the GitHub Marketplace under Code Quality / Continuous Integration.