websearch
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from an external search provider, creating a standard surface for indirect prompt injection via search results.
- Ingestion points: Search results, snippets, and knowledge graph data are fetched from the Serper API in
src/websearch/websearch.py. - Boundary markers: The output is formatted with result indices and separators, but lacks explicit instructions for the agent to ignore potentially malicious instructions embedded in search snippets.
- Capability inventory: The skill is limited to performing network requests to the Serper API and reading the specific configuration file
~/.prime/agent/auth.jsonfor credentials. - Sanitization: The implementation performs basic whitespace stripping on external content but does not filter for control characters or hidden injection strings.
Audit Metadata