Guard · CLAUDE.md scorecard

CLAUDE.md scorecard

Does your CLAUDE.md teach the load-bearing things a Claude Code session needs to know? One row per rule, a score, and the exact fix when a rule is missing.

Where to find it

  • Localhost: /scorecard.html?repo=<id>
  • API: GET /api/scorecard?repo=<id>
  • Keyboard: ⌘ K then scorecard
  • Sidebar: Brain → Context health → CLAUDE.md scorecard

What it does for you

A score, banded by color.The orb shows your score out of the baseline. Green at 85 and up, amber from 60 to 84, red below 60. The line under it reads how many rules pass and how long your CLAUDE.md is.
One row per rule, pass or fail.Each of the load-bearing CLAUDE.md clauses (worktree rule, three build tracks, twin gate, secrets, simplicity-first, and the rest, about 21 in all) gets a row: a check or cross, its severity (Error, Warn, or Info), the rule id, and a message.
A fix hint on every miss.A failed rule names the gap and shows the exact thing to add to your CLAUDE.md, so you can close it without guessing.
Missing-file check.If no CLAUDE.md exists at the expected path, the card says so instead of scoring an empty file.

Use it well

Tick Show only failing rules to see just the gaps. Rules come from plain files under lib/rules/claudemd.*.mjs, so a new rule appears here on its own. This pairs with the Context health score, which uses the same shared severity-to-deduction weights.

Read more