harden
Pass
Audited by Gen Agent Trust Hub on Jul 6, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection when analyzing external data.
- Ingestion points: The subagent reads file contents, git diffs, and project documentation (ADRs, reviews) using the Read, Grep, and Bash tools as instructed in SKILL.md and agent-prompt.md.
- Boundary markers: The prompt template in agent-prompt.md does not use delimiters or explicit instructions to treat the analyzed code as untrusted data, increasing the risk that the subagent follows instructions embedded within the code.
- Capability inventory: The subagent has access to Bash, Write, and Read tools, and can be granted Edit permissions for approved fixes, providing a significant surface for unauthorized actions if an injection occurs.
- Sanitization: There is no evidence of sanitization, filtering, or validation of the code or documentation before it is processed by the subagent.
Audit Metadata