deep-research

Pass

Audited by Gen Agent Trust Hub on Aug 15, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes content from untrusted external sources via the web_extract tool, which creates a surface for indirect prompt injection.
  • Ingestion points: External web content retrieved during the iterative loop (Step 3b in SKILL.md).
  • Boundary markers: The skill implements a 'Quality Filter' (Step 3c) and a 'Structured Evidence' format (references/structured-evidence-format.md) which separates verbatim snippets from synthesized analysis. It also uses explicit numbered citations.
  • Capability inventory: Includes web_search, web_extract, execute_code (for financial modeling), and delegate_task (for subagent management).
  • Sanitization: Employs a multi-step 'Iterative Loop' with an explicit 'Stopping Check' (Step 3f) and a 'Synthesis' phase that requires corroboration for facts to be labeled as [VERIFIED].
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 15, 2026, 06:16 PM
Security Audit — agent-trust-hub — deep-research