garden
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill operates by scanning untrusted data (the user's codebase) and performing write actions (modifying code and creating pull requests) based on that data. This creates a surface for indirect prompt injection if the scanned code contains instructions intended to mislead the agent.
- Ingestion points: The skill reads from codebase files during the 'Lenses' and 'Scan' phases to identify drift (SKILL.md).
- Boundary markers: There are no instructions to the agent to use delimiters or to ignore instructions embedded within the codebase files it reads.
- Capability inventory: The skill is authorized to modify files, create new git branches, and open pull requests (SKILL.md).
- Sanitization: The process does not include a step to sanitize or validate the content of the findings before they are used to generate code fixes.
Audit Metadata