review
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses local Node.js scripts (
contract-snapshot.mjsandscan-diff-hygiene.mjs) to perform repository analysis. These scripts execute standardgitcommands and file system operations to compare repository states and identify code hygiene issues. - [COMMAND_EXECUTION]: The skill instructions (e.g., in
references/verification-map.md) define a workflow for executing repository-specific validation commands such aspnpm compileandpnpm build. These are gated by 'Trust Gates' that require higher scrutiny for untrusted forks and explicit user authorization for any write operations. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from pull request bodies, comments, and diffs. This attack surface is mitigated by comprehensive instructions in
SKILL.md(Global Invariants) which mandate that such content must be treated strictly as data for review and never as instructions to the agent. - [DATA_EXF_EXPOSURE]: The skill defines strict security boundaries in
references/redlines.md(Redline 9), explicitly forbidding the recording, logging, or persistence of credentials, tokens, or API keys, and requiring path normalization to prevent directory traversal attacks. - [SAFE]: All identified command execution capabilities and data processing behaviors are well-documented, constrained by project-specific security rules, and aligned with the primary purpose of a developer-focused code review tool.
Audit Metadata