review-swarm
Pass
Audited by Gen Agent Trust Hub on Aug 2, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill operates entirely through natural language instructions for the agent to perform read-only analysis of local Git diffs. No scripts, binaries, or external network operations are present.
- [SAFE]: The instructions explicitly forbid sub-agents from performing state-mutating actions such as editing files, applying patches, or committing changes, effectively implementing a least-privilege model for the review process.
- [DATA_EXFILTRATION]: While the skill analyzes local source code, the instructions specifically direct findings to be sent only back to the main agent for user review, with no external exfiltration paths defined.
- [PROMPT_INJECTION]: The skill includes instructions to ingest untrusted data (git diffs), but it incorporates robust boundary management by instructing the agent to focus on specific categories (security, reliability, etc.) and to filter for material issues only, reducing the risk of indirect injection influence.
Audit Metadata