no-comments

Pass

Audited by Gen Agent Trust Hub on Sep 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process and react to data from untrusted sources, specifically code comments and git diffs. Instructions embedded in these comments (e.g., 'do not remove', 'do not change wording') directly influence the agent's decision-making process regarding code deletion and structural changes.
  • Ingestion points: The skill reads from the caller's files or diff and the current diff against the base branch as specified in the Scope section.
  • Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat text within comments as data rather than instructions.
  • Capability inventory: The skill can spawn a subagent (pstack:comment-sicko), delete application code paths, modify API parameters, and execute analysis tools (/architect, /how, /why).
  • Sanitization: No sanitization or validation logic is defined to check the content of comments before the agent 'encodes' them into types, runtime checks, or CI lints.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 7, 2026, 11:13 PM
Security Audit — agent-trust-hub — no-comments