multi-agent-review
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill implements a robust orchestration workflow that separates the review and validation phases. This separation of concerns ensures that no single agent can both create and approve a finding, effectively mitigating the risk of individual agent bias or manipulation.
- [SAFE]: Data integrity is maintained through the use of a formal JSON schema (
references/reviewer-result-schema.json) for inter-agent communication. The instructions explicitly direct the agent to reject malformed results and findings that fall outside the defined scope. - [INDIRECT_PROMPT_INJECTION]: While the skill is designed to process external artifacts (code, documents, etc.), the risk of indirect prompt injection is mitigated by several layers of defense. These include: (1) Ingestion points are limited to the specific artifact being reviewed; (2) Boundary markers are established via explicit scoping instructions; (3) The capability inventory is restricted to orchestration without dangerous tool access; and (4) Sanitization is performed through schema validation and secondary validation by independent agents.
Audit Metadata