Score your CLAUDE.md or AGENTS.md.

Paste an instruction file and get its footprint score with the top fixes. It uses the same scoring as rig doctor --lite. The text is scored in memory and thrown away: it is not stored or logged.

What it checks

• Size: one point per 250 estimated tokens past 3,500 per engine, up to 30. Tokens are estimated as characters divided by four.
• Dated lines: one point each, up to 10.
• SUPERSEDED, DEPRECATED, RETIRED markers: two points each, up to 10.
• Em and en dashes: half a point each, up to 5.
• Repeated lines over 40 characters: one point each, up to 10.
• Secret-shaped values: ten points each, up to 30.
• Absolute home paths: half a point each, up to 5.

Grades: A from 90, B from 75, C from 60, D from 40. A score describes the footprint, not whether the rules are right. How to bring a score up.

For agents

POST https://commonrig.com/v1/score
Content-Type: application/json

{"claude_md": "...", "agents_md": "..."}

The JSON answer has the score, grade, deductions and top fixes. Both endpoints are rate limited. A2A agents can send the same fields as a data part to https://commonrig.com/a2a.

Score the whole setup

A paste sees one file per engine. The free sample's rig doctor --lite reads the files your engines actually load. The hosted doctor check (A$0.50 a run) scores a masked bundle of your setup, including whether the hooks your policy expects are registered, and suits CI. The kit (A$10 once) gives you the rule source, generator and hooks to keep the score up.