deep-research
Pass
Audited by Gen Agent Trust Hub on Aug 9, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill contains no executable code, remote dependencies, or network operations. It is composed entirely of instructional text guiding the agent on how to perform deep research.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, untrusted data from multiple sources. While this presents a standard attack surface for indirect prompt injection, the instructions provide strong mitigation strategies by mandating a 'skeptical' posture, source credibility scoring, and adversarial verification of claims. No specific tool-calling vulnerabilities were identified.
- Ingestion points: The skill explicitly instructs the agent to 'fan out across sources' and 'go to primary sources' (SKILL.md).
- Boundary markers: Includes a 'Source-credibility checklist' and instructions to 'treat every key claim as guilty until corroborated'.
- Capability inventory: No scripts provided; behavior depends on the agent's native search and retrieval tools.
- Sanitization: Relies on logical verification and source scoring rather than technical input filtering.
Audit Metadata