requesting-code-review
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions and prompt templates involve the execution of standard git commands (
git rev-parse,git log,git diff,git show) to analyze repository history and changes. These are typical operations for a code review tool.\n- [INDIRECT_PROMPT_INJECTION]: The subagent defined incode-reviewer.mdprocesses potentially untrusted data from code diffs and plan requirements.\n - Ingestion points: The subagent prompt incorporates variable content including
[DESCRIPTION],[PLAN_OR_REQUIREMENTS], and the output ofgitcommands incode-reviewer.md.\n - Boundary markers: The template does not utilize specific delimiters to isolate untrusted content from the instructions.\n
- Capability inventory: The subagent is authorized to use git inspection tools and create temporary worktrees using
git worktree add.\n - Sanitization: No explicit sanitization or filtering of the ingested data is performed.
Audit Metadata