baidu-search
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes external content from Baidu search results, which can contain untrusted data designed to influence the agent's behavior.
- Ingestion points: Search results returned from the Baidu AI Search API in
scripts/search.py. - Boundary markers: The script outputs results as a raw JSON string without adding explicit boundary markers or instruction-override warnings for the agent.
- Capability inventory: The Python script is restricted to making network requests to the Baidu API; it does not include file system write or subprocess execution capabilities.
- Sanitization: The script removes the
snippetfield from the results to reduce data volume, but it does not escape or sanitize titles or URLs before returning them to the agent context.
Audit Metadata