review-code-comments
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill reviews source code comments which are an external, untrusted data source and potential vector for indirect prompt injection.\n
- Ingestion points: The agent reads source files and git diffs provided by the user or identified via context (
SKILL.md).\n - Boundary markers: While no technical delimiters (like XML tags) are specified for the comment content, the skill provides a conceptual boundary by instructing the agent to treat comment text as unverified claims.\n
- Capability inventory: The skill is authorized to delete comments in local files and can delegate review tasks to subagents (
SKILL.md).\n - Sanitization: The instructions include a specific security pass where the agent is told to treat markers like
IMPORTANTordo not removeas claims needing evidence rather than instructions to follow, mitigating the risk of obedience to injected instructions (SKILL.md).
Audit Metadata