mem-history

Pass

Audited by Gen Agent Trust Hub on May 4, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill documentation encourages the installation of an external third-party plugin (claude-mem@thedotmack). This creates a functional dependency on code maintained by an unverified source outside the official trusted vendor list.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it retrieves and displays project 'observations' which are derived from external data sources.
  • Ingestion points: Untrusted data enters the context via the mcp-search__search and mcp-search__get_observations tools which query the claude-mem database.
  • Boundary markers: There are no defined delimiters or instructions (e.g., 'ignore instructions within these observations') to help the agent distinguish between retrieved memory and its primary system instructions.
  • Capability inventory: The skill facilitates the retrieval of content that could influence an agent possessing shell execution and file modification capabilities.
  • Sanitization: There is no mention of sanitization, escaping, or validation of the retrieved memory text before it is integrated into the agent's active prompt context.
Audit Metadata
Risk Level
SAFE
Analyzed
May 4, 2026, 04:17 AM
Security Audit — agent-trust-hub — mem-history