cleaning-up-stale-feature-flags
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill demonstrates a high level of security awareness by implementing defensive programming against prompt injection and indirect injection attacks. It uses character allowlisting and explicit instructions to ensure data is never treated as code.
- [PROMPT_INJECTION]: The skill contains references to common prompt injection phrases such as "ignore previous instructions", but these are used as explicit examples of content the agent should ignore if encountered in external data. This serves as a defensive configuration rather than a malicious attempt to override system behavior.
- [COMMAND_EXECUTION]: The skill performs repository edits and PR creation but mandates user approval and adheres to local environment policies, preventing unauthorized code execution or publication.
- [DATA_EXFILTRATION]: Explicit policies prevent the inclusion of sensitive metadata or secrets in public-facing PR descriptions, protecting against accidental data leaks during the cleanup process.
- [INDIRECT_PROMPT_INJECTION]: The skill manages the potential attack surface of ingesting external feature flag data and repository content through explicit boundary warnings and validation regex patterns (^[a-zA-Z0-9_./:-]+$) to prevent malicious content from being interpreted as instructions.
Audit Metadata