cognee-community

Pass

Audited by Gen Agent Trust Hub on Aug 18, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill documents the use of connectors to ingest data from external platforms such as Slack, Gmail, Notion, and Google Drive. This creates a surface where the agent might process untrusted data containing embedded instructions.
  • Ingestion points: Data is ingested via community connectors (e.g., slack_export_source) as described in SKILL.md.
  • Boundary markers: The provided examples do not demonstrate the use of boundary markers or instructions to ignore embedded commands within the ingested data.
  • Capability inventory: Ingested data is processed and stored in graph and vector databases, which the agent may subsequently retrieve and act upon.
  • Sanitization: The skill documentation does not mention specific sanitization or filtering protocols for external content before it is stored in the memory engine.
  • [EXTERNAL_DOWNLOADS]: The documentation guides users to install external packages from PyPI to extend the framework's functionality.
  • Evidence: Examples include uv pip install cognee-community-vector-adapter-qdrant. These packages are official community extensions managed by the skill's author and represent the intended use of the framework.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 18, 2026, 03:33 AM
Security Audit — agent-trust-hub — cognee-community