review-file
Pass
Audited by Gen Agent Trust Hub on Aug 17, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze the full content of external files provided by the user, creating a surface for indirect prompt injection attacks. Malicious instructions hidden within the target file (e.g., in comments or string literals) could attempt to override the reviewer's logic or bypass security checks.
- Ingestion points: The agent reads the entire named file during the 'Orient' pass.
- Boundary markers: The instructions do not define any boundary markers or delimiters (such as specific XML tags or start/end markers) to prevent the agent from confusing untrusted data with its primary instructions.
- Capability inventory: The agent has the capability to read local files, fetch external API documentation, and perform file edits in its 'Apply mode'.
- Sanitization: There is no requirement for the agent to sanitize, escape, or validate the content of the file before processing it.
- [COMMAND_EXECUTION]: In the 'Apply mode', the agent is authorized to modify the file system by rewriting parts of the reviewed file. While this functionality requires explicit user confirmation (e.g., 'apply' or 'do 2'), a successful indirect prompt injection could lead the agent to propose and execute harmful modifications that the user might inadvertently approve.
Audit Metadata