memory-systems

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill references official and well-known GitHub repositories for memory management tools, including Graphiti (github.com/getzep/graphiti) and Cognee (github.com/topoteretes/cognee). These references are for documentation and legitimate software distribution purposes.\n- [PROMPT_INJECTION]: The memory ingestion patterns described (e.g., using cognee.add) identify a surface for indirect prompt injection where external data is incorporated into the agent's knowledge base.\n
  • Ingestion points: The skill demonstrates adding files and raw data using cognee.add() and cognee.cognify() in SKILL.md.\n
  • Boundary markers: Delimiters or 'ignore instructions' warnings are not present in the provided snippets.\n
  • Capability inventory: Retrieved memory content influences agent reasoning and context generation.\n
  • Sanitization: The architecture overview does not specify content sanitization or validation steps during the ingestion phase.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 01:40 AM