review
Pass
Audited by Gen Agent Trust Hub on Jul 6, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection as it processes untrusted source code and repository diffs. This data is provided to a subagent which has access to tools like Bash and Write. Ingestion points: The skill reads git diff output and repository files using the Read tool. Boundary markers: No explicit delimiters or instructions are used in the subagent prompt to separate the untrusted code from the agent's instructions. Capability inventory: The subagent is granted tools including Bash, Write, Read, Grep, and Glob. Sanitization: Untrusted code content is interpolated directly into the subagent prompt without escaping or sanitization.
- [COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute git commands for branch analysis and change detection. Evidence: The skill runs git rev-parse, git merge-base, git diff, and git ls-files to identify the files and changes to be reviewed.
Audit Metadata