cognee-recall
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The
recall()API documented in this skill serves as an entry point for processing untrusted user queries against a knowledge graph. This creates an attack surface for indirect prompt injection, where user-supplied strings influence the retrieved context passed to an LLM. - Ingestion points: The
cognee.recall()method processes natural language query strings as seen inSKILL.mdexamples. - Boundary markers: Not explicitly detailed in the provided documentation snippets.
- Capability inventory: The skill facilitates graph search, raw Cypher execution (including writes), and multi-step agentic loops via
AGENTIC_COMPLETION. - Sanitization: The documentation highlights a permission layer (handling
PermissionDeniedErrorandDatasetNotFoundError) that restricts data access based on user grants. - [COMMAND_EXECUTION]: The documentation describes the
CYPHERsearch type, which enables the execution of graph database queries. The skill includes a clear security warning that this feature is capable of write operations and should be restricted to trusted queries. It also details theNATURAL_LANGUAGEtype, which involves AI-generated query execution.
Audit Metadata