source-deep-dive
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves fetching content from external URLs using the
WebFetchtool and processing that data to build an evidence base. This constitutes a vulnerability surface where malicious instructions hidden in external web pages (e.g., in HTML comments or invisible text) could potentially influence the agent's extraction or validation logic. - Ingestion points: Data is ingested from external websites via
WebFetchand from the source map filetmp/research/_sources-{slug}.md. - Capability inventory: The skill utilizes
WebFetch, has the authority to spawn sub-agents (spawn_agent), and performs file read/write operations within thetmp/anddocs/directories. - Boundary markers: The skill instructions emphasize critical thinking, skepticism, and mandatory cross-validation (requiring 2+ independent sources for high confidence). These logical constraints help mitigate accidental obedience to injected instructions.
- Sanitization: The instructions do not specify technical sanitization or escaping of the raw content fetched from the web before processing.
Audit Metadata