Scoring

A 0–100 agent-readiness score across five weighted categories. Same rubric for every repo; open-source rules.

Discoverability — 25 points

Does an agent dropped into the repo know what it is and where things live? README quality, AGENTS.md presence and content, repo metadata.

Buildability — 25 points

Can the agent run the project without asking? Lockfile, declared scripts (build/test/lint/typecheck), env template, CI that uses the same commands.

Conventions — 20 points

Are the rules of the road written down? Commit format, linter and formatter configs, editorconfig, documented naming and folder structure.

Documentation — 15 points

License, contributing guide, code of conduct — and when you pass --url, the public docs surface gets audited too.

Safety — 15 points

Secrets out of the tree, SECURITY.md, pre-commit hooks, PR template, CODEOWNERS, declared off-limits zones for agents.

Renormalization

Not every category applies to every repo — a repo with no public docs site skips the documentation-surface rules. The score is renormalized against the categories that apply, so it's always 0–100 and always comparable:

score = round(100 × earned / possible_applicable)

Exit codes

0   score ≥ 80
1   score < 80 (local CI gate)
2   org policy violated (Pro plan, --push only)