cognee-improve-sessions
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents functionality that ingests untrusted user content and processes it via an LLM for knowledge graph enrichment.
- Ingestion points: The skill describes storing conversation turns, agent traces, and feedback using functions like
cognee.remember()andcognee.recall()(SKILL.md). - Boundary markers: No instructions are provided for implementing boundary markers or using specific delimiters to separate user-provided data from system instructions during distillation.
- Capability inventory: The
cognee.improve()operation includes multiple stages, such asdistill_sessionsandglobal_context_index, which utilize an LLM to summarize session data and incorporate it into a permanent knowledge graph (SKILL.md). - Sanitization: The documentation does not specify any sanitization, filtering, or validation logic to be applied to session data before it is processed by the LLM-driven enrichment stages.
Audit Metadata