private-company-research
Pass
Audited by Gen Agent Trust Hub on Jul 14, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is designed to ingest data from external sources, creating an indirect prompt injection surface.
- Ingestion points: Data is gathered via web searches and references to external platforms such as LinkedIn, Weibo, and Reddit (SKILL.md).
- Boundary markers: No formal technical delimiters are defined for external data, though the skill requires labeling every point with confidence markers (SKILL.md).
- Capability inventory: The skill has the capability to perform web searches, coordinate multiple agents via run_swarm, and write reports to the local file system (SKILL.md).
- Sanitization: Mitigation is implemented through logic-based sanitization involving mandatory data conflict arbitration and signal consistency checks.
Audit Metadata