agent-friendly-action

GitHub Action that posts a per-PR delta of the Agent Friendly Code score — "this PR drops your Claude Code score by 4.1 points because it removed CI config."

Version: v0.1.1 — plans for the initial cut live in tasks/0.1.0/. Pin @v0 to follow the latest 0.x release; pin a precise tag (@v0.1.1) if you want to opt out of automatic minor/patch updates. Tracking task in the parent project: tasks/0.5.0/02-score-diff-on-pr.md.

What it does

On every pull request, the action:

1. Reads the PR head and base refs from the GitHub event payload.
2. Fetches the base ref locally (no network clone — actions/checkout already gave us the head).
3. Scores both trees with the bundled scorer.
4. Posts (or edits) a single PR comment with the delta + per-signal breakdown.

It runs entirely inside your CI — no third-party server is required. The bundled scorer is the only thing that decides the delta; the Agent Friendly Code web app is not contacted at runtime.

Install

# .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 }}

AGENTS_BADGE_TOKEN is the opt-in marker, not an API credential. Set it in the repo's secrets to enable the comment; leave it unset and the action exits silently.

Inputs

Input	Required	Default	Description
agents-badge-token	yes	—	Opt-in marker. Action no-ops when empty.
github-token	no	${{ github.token }}	Used to post / edit the PR comment.
base-ref	no	event base	Override the base ref (advanced — usually unneeded).

Why opt-in?

Two reasons:

1. Templates / starter repos can ship the workflow file without it firing for forks or unenrolled users — only repos that explicitly set the secret get the comment.
2. Future-proofing — when the upstream web app exposes a runtime weights endpoint (deferred to its 1.0.0/03 benchmark harness), the same env var can be promoted to a real key with no workflow change.

Self-contained

The scorer is bundled into dist/ via @vercel/ncc. The action does not call the Agent Friendly Code web app at runtime — weights and signal logic ship inside the bundle. If the web app goes offline tomorrow, this action keeps producing PR comments unchanged.

Want this in your editor too?

The agent-friendly-skill sibling vendors the same scorer for your editor. Profiles the same 8 agents this action does (Claude Code, Cursor, Devin, GPT-5 Codex, Gemini CLI, Aider, OpenHands, Pi); installs into any vercel-labs/skills-compatible agent. Recommendation is score-driven — the skill picks the highest-scoring profile as the best fit and tells you which model class to switch to; you do the actual switch.

npx skills add hsnice16/agent-friendly-skill

Sponsor

If this action is useful to you, consider sponsoring its development: github.com/sponsors/hsnice16.

License

MIT