multi-source-search

Pass

Audited by Gen Agent Trust Hub on Sep 1, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill requires the agent to ingest and process content from various web sources, creating a potential surface for indirect prompt injection.\n
  • Ingestion points: External data is retrieved via the agent's native search and page-reading tools (SKILL.md).\n
  • Boundary markers: The instructions include explicit warnings to ignore instructions embedded in search results and to treat all retrieved content as untrusted evidence (SKILL.md, Overview).\n
  • Capability inventory: The skill uses a bundled Python script, scripts/validate_report.py, which is executed locally to perform schema validation on JSON files.\n
  • Sanitization: The Python validator script ensures structural correctness of the report and canonicalizes URLs, but does not perform natural language filtering of the claim content.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 1, 2026, 07:34 AM
Security Audit — agent-trust-hub — multi-source-search