seo-topic-research-pipeline

Pass

Audited by Gen Agent Trust Hub on Jul 27, 2026

Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it is designed to fetch and process data from third-party sources.
  • Ingestion points: Website content is retrieved via WebFetch in Phase 1A; Sales call transcripts and summaries are retrieved from Fathom or Fireflies MCP servers in Phase 1C and Phase 1D.
  • Boundary markers: The skill lacks explicit boundary markers or instructions to the agent to disregard potential instructions embedded within the fetched external content.
  • Capability inventory: Across the pipeline, the agent has the capability to perform network requests (WebFetch) and write to the local file system (Phase 8).
  • Sanitization: There are no described mechanisms for sanitizing or filtering instructions from the external inputs before they are processed by the agent.
  • [DATA_EXFILTRATION]: The skill interacts with sensitive business information and non-whitelisted external services.
  • Sensitive Data Access: The skill is configured to ingest sales call transcripts (via Fathom/Fireflies) and search performance data (via Google Search Console), which contain proprietary business intelligence.
  • Network Operations: The skill performs network operations to external services including Ahrefs and general web domains (via WebFetch). While these are aligned with the skill's primary purpose, they represent a potential data egress point if the agent is manipulated via indirect injection.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 27, 2026, 09:06 AM
Security Audit — agent-trust-hub — seo-topic-research-pipeline