desearch-web-search

Pass

Audited by Gen Agent Trust Hub on Mar 7, 2026

Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill ingests untrusted search results from the web, creating an indirect prompt injection surface.\n
  • Ingestion points: Data is ingested from the Desearch API via the api_request function in scripts/desearch.py.\n
  • Boundary markers: Absent. The script does not utilize explicit delimiters or instructions to frame the retrieved content for the agent.\n
  • Capability inventory: Across all scripts, capabilities are limited to network requests to api.desearch.ai. There is no file-system write access or subprocess execution capability.\n
  • Sanitization: Absent. No filtering or sanitization of snippets or titles is performed before they are returned to the agent.\n- [DATA_EXFILTRATION]: The skill connects to the vendor's API to perform its core search function.\n
  • Evidence: Network requests are made to https://api.desearch.ai. This is a vendor-owned resource used as intended by the skill's author 'Desearch-ai'.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 7, 2026, 11:54 AM