no-comments
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted source code and evaluates instructions and claims found within comments to decide how to refactor the code. This creates a vulnerability to indirect prompt injection where malicious comments could manipulate the agent's logic during the refactoring phase.
- Ingestion points: The agent ingests 'the caller's files or diff' and the 'working tree' (SKILL.md).
- Boundary markers: Absent. There are no instructions for the agent to use delimiters or specific safety protocols to isolate code content from its own instruction processing.
- Capability inventory: The agent is authorized to delete comments, modify file content, delete dead paths, drop parameters, and implement root-cause fixes (SKILL.md).
- Sanitization: None. Although a 'read-only subagent' is suggested for the initial review, the main agent acts upon that review to modify files, and no sanitization of the comment content is performed.
- [NO_CODE]: The skill does not include any executable scripts, binaries, or source code; it consists solely of markdown instructions and a license file.
Audit Metadata