source-deep-dive
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill uses a
WebFetchtool to retrieve content from external URLs identified in a prior source map. It is capped at 8 calls per invocation. - [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from external websites, which could potentially contain malicious instructions designed to influence the agent's behavior.
- Ingestion points: Data retrieved from external URLs via the
WebFetchtool (SKILL.md, Step 2). - Boundary markers: The instructions do not define explicit boundary markers or XML-style tags to isolate the fetched external content from the agent's internal instructions.
- Capability inventory: The skill has network access capabilities via
WebFetchand file system access (read/write) within thetmp/research/directory. - Sanitization: There is no explicit mention of sanitization or filtering for the fetched web content; the skill relies on the agent's ability to extract specific data points (claims, quotes, methodology) accurately.
Audit Metadata