Your agents read this file 200 times a day. Score it once.
Paste your CLAUDE.md or AGENTS.md. Twelve checks, a score out of 100, and the three lines that matter most. No account, no email, nothing kept.
The text stays in memory for the request. Nothing is written to a database or sent anywhere.
Why these twelve
Each check maps to a failure we have seen break a real session, not a theory. The order matters: the first four decide whether an agent can be trusted with anything, the next four decide whether it can work with other people, the last four decide whether it keeps working next month.
- Boundary, gates, definition of done, status format.
Without these four, the agent reports confidence instead of work. This is the "done but nothing changed" problem.
- Secrets, precedence, question policy, output path.
These are the team rules. They matter the moment two people or two sessions share the folder.
- Money priority, context budget, placeholders, honesty.
These keep the file useful: it aims at revenue, stays short enough to be read on every task, and stops invented numbers.
Most files score 40–55 on the first run. The usual three misses: no explicit gate list, no three-line status format, and a file that has grown past 1,200 words without a structure.
Want the whole thing done for you? $29 context fix in 24 hours, or $299 audit for a team.