deep-research
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to perform multi-source, parallel research ('fan-out') and mandates that all findings be written to durable files on disk. This process ingests untrusted content from the internet and places it directly into the local environment.
- Ingestion points: Parallel research subagents gathering data from distinct angles, companies, and source classes via external investigations (SKILL.md).
- Boundary markers: None specified. The instructions emphasize capturing content and URLs 'verbatim' without recommending delimiters or 'ignore' instructions for the subagents processing the data.
- Capability inventory: The skill utilizes file-writing capabilities to save synthesis and appendix documents to
research/andstartup/directories. - Sanitization: None specified. The instructions focus on detail preservation (capturing 'raw numbers', 'verbatim source URLs') rather than filtering or sanitizing the ingested external data before it is written to the file system.
Audit Metadata