requesting-code-review
Pass
Audited by Gen Agent Trust Hub on Jun 21, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill defines a subagent prompt that ingests untrusted data from task descriptions and requirements documents, creating a surface for indirect prompt injection.
- Ingestion points: External implementation descriptions and requirement plans in code-reviewer.md via {DESCRIPTION} and {PLAN_OR_REQUIREMENTS}.
- Boundary markers: Uses Markdown headers (## What Was Implemented, ## Requirements / Plan) but lacks explicit delimiters instructing the agent to ignore embedded instructions within the variables.
- Capability inventory: The subagent is instructed to run git diff and git diff --stat to inspect code changes.
- Sanitization: No sanitization or validation of the plan or description content is performed before interpolation.
- [COMMAND_EXECUTION]: The skill uses shell-based git commands to identify commit ranges and analyze code differences. These operations are standard for development tools and are restricted to repository metadata and diffs.
Audit Metadata