decomplex
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from project files (source, tests, config), which creates an attack surface for indirect prompt injection.
- Ingestion points: As specified in
SKILL.md, the agent reads "reviewed targets" provided by the user. - Boundary markers: The skill contains "Hard rules" that explicitly restrict output to the
.reviews/directory and forbid editing any reviewed artifacts, creating a clear operational boundary. - Capability inventory: The skill is limited to file system read access for targets and write access for a single report file; it lacks network access, shell execution, or privilege escalation capabilities.
- Sanitization: The skill mandates a human-in-the-loop workflow, instructing the agent to never auto-implement recommendations and to ask the user when ambiguity arises, which prevents the autonomous execution of potentially malicious instructions found in analyzed data.
Audit Metadata