AgentDB Memory Patterns

Pass

Audited by Gen Agent Trust Hub on Jul 12, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the user to download and execute the AgentDB CLI using npx agentdb@latest. This is a standard method for running NPM-based tools without local installation.
  • [REMOTE_CODE_EXECUTION]: The use of npx agentdb@latest results in the execution of code downloaded from the NPM registry at runtime.
  • [COMMAND_EXECUTION]: The skill provides numerous shell commands for database initialization (init), querying (query), and plugin management (create-plugin). It also includes a command to add an Model Context Protocol (MCP) server to Claude Code using claude mcp add agentdb npx agentdb@latest mcp.
  • [INDIRECT_PROMPT_INJECTION]: The skill implements a memory system that ingests and stores interaction data (e.g., storeMessage, insertPattern). This creates an attack surface where untrusted user input could be stored and later retrieved into the agent's context.
  • Ingestion points: Data enters the system via API calls like adapter.insertPattern, db.storeMemory, and db.storeFact in SKILL.md.
  • Boundary markers: The provided code snippets do not explicitly show boundary markers or 'ignore' instructions for the stored data.
  • Capability inventory: The skill includes file system operations (via the agentdb CLI interacting with .db files) and network operations (implied by the NPM package downloads and MCP server functionality).
  • Sanitization: There is no evidence of sanitization or filtering for the data stored in the memory patterns.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 12, 2026, 11:25 PM
Security Audit — agent-trust-hub — AgentDB Memory Patterns