A small Python script that reads the rule files your AI coding assistant uses and reports problems in how they are written. It only reads files. It never edits, uploads or sends anything.

What you get (18 files in one zip)

- contextos_audit.py: the auditor, standard library only

- README and CHANGELOG

- audit-prompt.md: a prompt that asks your assistant to review what your rules mean

- Two example projects (one deliberately broken, one clean) with saved outputs

What it checks (15 checks)

Files that are too long, plain .md files inside .cursor/rules (Cursor ignores them), .mdc rules without front matter, alwaysApply together with globs, invalid paths: globs, vague instructions like "write clean code", possible contradictions, duplicate lines, missing test or lint commands, unfilled placeholders, secrets inside rule files, .cursorrules files, broken @path imports, unreadable files.

How to run

python3 contextos_audit.py path/to/your/project

Options: --json, --strict (exit code 1 on warnings, for CI), --no-color.

Status and limits

- Tested on Python 3.11, 3.12 and 3.13 with identical results, on the two example projects and four edge cases (empty folder, missing folder, JSON output, duplicate detection). Older Python versions, Windows paths and very large repositories are untested.

- It checks wording and structure. It does not measure whether a model follows your rules.

- Conflict, vagueness and duplicate checks are heuristics: expect some misses and some false alarms.

- Checks are based on the official Cursor and Claude Code documentation read on 2026-10-19.

License: personal and commercial use in your own projects. Do not resell or redistribute.