contract-stress-test
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted contract drafts (PDF/DOCX) which could contain adversarial content intended to influence agent reasoning or bypass review criteria.
- Ingestion points: Contract documents and prior review artifacts (SKILL.md lines 114, 150).
- Boundary markers: Employs mandatory disclaimer blocks and metadata structures to delimit contract data from instructions (SKILL.md lines 162-184).
- Capability inventory: The skill does not utilize network operations, shell execution, or persistent file storage, minimizing the impact of potential injection.
- Sanitization: Implements a scope boundary that rejects hallucinated facts and requires a named human tiebreaker for all high-severity or contested findings (SKILL.md lines 231-255).
Audit Metadata