systematic-debugging
Warn
Audited by Gen Agent Trust Hub on Mar 8, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The file
find-polluter.shprocesses filenames using a shellforloop and word splitting on the output offind. This can lead to unexpected behavior or execution errors when handling files with spaces or special characters in their names. - [CREDENTIALS_UNSAFE]: The
SKILL.mdfile contains debugging examples that suggest revealing the state of secrets in logs and querying the system keychain usingsecurity list-keychainsandsecurity find-identity. These actions can lead to the exposure of sensitive cryptographic identities and credentials during the debugging process. - [PROMPT_INJECTION]: The skill employs authoritative framing, such as 'The Iron Law' and 'The Four Phases', combined with absolute directives like 'You MUST' and 'Violating... is violating the spirit'. This instructional style is designed to override the agent's default reasoning processes, which could be leveraged to bypass safety filters or ignore standard constraints under the guise of debugging requirements.
Audit Metadata