skills/radimsem/remindb/memorize/Gen Agent Trust Hub

memorize

Pass

Audited by Gen Agent Trust Hub on Jun 15, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill establishes a workflow for ingesting untrusted data from local files or user input into a long-term memory database, creating a surface for indirect prompt injection.
  • Ingestion points: Data is brought into the system via files in the directory defined by the $REMINDB_SOURCE environment variable (processed by MemoryCompile) or through direct text input via the payload argument in MemoryWrite tools.
  • Boundary markers: There are no instructions or templates provided that use unique delimiters or explicit "ignore embedded instructions" warnings when processing or retrieving these memories.
  • Capability inventory: The skill leverages several high-capability tools including MemoryWrite (DB modification), MemoryCompile (file system processing), MemoryForget (data deletion), and MemoryRollback (state restoration).
  • Sanitization: No mechanisms for escaping, filtering, or validating the content of the stored data are mentioned to prevent malicious instructions from being treated as authoritative commands upon retrieval.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 15, 2026, 02:22 AM
Security Audit — agent-trust-hub — memorize