Paste a CLAUDE.md or AGENTS.md file to inspect silent failure patterns: truncation risk, missing evidence, vague completion, and rules that read like controls but cannot enforce themselves.
● Runs only in your browser. Nothing is uploaded or sent.
The text stays in this page. No server, analytics request, AI model, fetch, or XMLHttpRequest is used.
Paste a rules file and run the audit.
This page splits your text into lines and applies deterministic size, structure, and pattern checks. Each finding cites the relevant line and excerpt so you can decide whether it applies.
It looks for known weak signals, not intent. A pass means an applicable static check found no issue. If the required input structure is absent, the check is not evaluated; no verdict is not a pass.
It does not use an LLM, call an API, upload your text, inspect your repository, or claim to verify runtime loading. For that, check your runtime path and test a fresh read-only agent process.
No. The audit runs entirely in this browser. The page contains no fetch or XMLHttpRequest code, and pasted text is not sent anywhere.
No. It uses deterministic text, line, and structure checks only. It does not call an AI model.
A warning identifies a rule-file pattern that may be fragile. It is a prompt to inspect the cited line, not proof that an agent will fail.
Not evaluated means the input did not contain the structure that a check needs. It is not a pass, and it is excluded from the passed-check score.
No. Static text cannot prove runtime loading. Check the runtime path, file-size limit, and a fresh read-only process separately.
Paste any CLAUDE.md, AGENTS.md, or comparable Markdown instruction file.
Read the pattern library for evidence-first rule-file practices, or browse practical developer operations materials.
CLAUDE.md patterns on GitHub · Developer operations materials