cognee-remember
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a local shell script (
cognee-remember.sh) and a CLI tool (cognee-cli) to store data. User input is passed to these scripts via the$ARGUMENTSvariable, which can be a vector for command injection if the scripts do not properly sanitize the input.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input and files, storing them in a knowledge graph for later retrieval, creating a surface for injection attacks.\n - Ingestion points: Untrusted data enters the system through the
$ARGUMENTSvariable and files uploaded via the--fileflag inSKILL.md.\n - Boundary markers: The instructions lack delimiters or warnings to the agent to ignore instructions contained within the ingested data.\n
- Capability inventory: The skill can execute local shell scripts, read files from the filesystem, and perform network POST requests to the Cognee API.\n
- Sanitization: No explicit sanitization or validation of the input data is defined in the skill instructions.\n- [DATA_EXFILTRATION]: The skill transmits potentially sensitive data and file contents to an external Cognee server API. Although this is the intended purpose, it involves sending local information to a remote endpoint.
Audit Metadata