• Everything runs locally. The scripts in this repository make no network calls, collect no telemetry, and send nothing anywhere. They read the skill directory you point them at and print a report.
• Zero runtime dependencies. Python standard library only — there is no third-party supply chain to audit beyond Python itself (pytest is used for development testing only).
• One deliberate exception to "read-only": scripts/script_tester.py executes the Python scripts of the skill under test (with a timeout) in order to test them. Only run it against skill code you trust, exactly as you would before running that code yourself.

Report suspected vulnerabilities by opening a GitHub issue on this repository. Since this tool runs entirely locally and holds no user data, public reporting is acceptable; include the affected script, a reproduction, and the impact you see.

Only the latest commit on main is supported with fixes.