deep-research

Pass

Audited by Gen Agent Trust Hub on Aug 31, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch external content via web_search and fetch_url, and process user-provided materials via list_materials and read_material. This creates an attack surface where malicious instructions embedded in web pages or user documents could influence the agent's behavior.
  • Ingestion points: External data enters the context through fetch_url (up to 6 URLs) and web_search (up to 8 calls), as well as user materials managed by list_materials and read_material (SKILL.md, Steps 1, 3, 4, 5).
  • Boundary markers: The instructions do not specify explicit delimiters or 'ignore instructions' wrappers for the fetched content, though they do require maintaining a ledger (SKILL.md, Step 5).
  • Capability inventory: The skill has the capability to perform web searches, fetch URLs, and generate course content (SKILL.md, Step 9).
  • Sanitization: There is no mention of sanitization or filtering of external content before it is processed by the page generator (SKILL.md, Step 9).
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 31, 2026, 10:40 AM
Security Audit — agent-trust-hub — deep-research