denarrate
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes the
git diff --stagedcommand to retrieve source code changes for analysis, which is a standard procedure for development utilities. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from source code comments, creating a surface for potential indirect prompt injection.
- Ingestion points: Untrusted data is ingested via file paths, directory contents, and the output of the
git diffcommand as described inSKILL.md. - Boundary markers: The instructions do not specify any delimiters or safety prompts to prevent the agent from interpreting instructions potentially hidden within the code comments.
- Capability inventory: The agent has the capability to read and write files locally and execute
gitcommands. - Sanitization: There is no mention of sanitizing or escaping the comment content before it is evaluated by the agent.
Audit Metadata