deep-research

Pass

Audited by Gen Agent Trust Hub on Feb 27, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its reliance on untrusted external data sources.
  • Ingestion points: External data is fetched from Exa search tools (mcp__exa__web_search_exa, mcp__exa__get_code_context_exa), YouTube transcriptions (yt-transcribe), and general web searches.
  • Boundary markers: The skill documentation lacks explicit boundary markers or instructions for the agent to ignore potentially malicious embedded content within the research findings before they are processed by synthesis agents.
  • Capability inventory: The system can write research findings and reports to the local file system (scratchpad) and coordinate multiple sub-agents, which increases the potential surface area for an indirect injection to influence the final output or agent behavior.
  • Sanitization: No sanitization or verification process for the content of the retrieved findings is described before the data is written to files or used as input for the synthesizer agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 27, 2026, 12:25 PM