AgentDB Memory Patterns
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructions and documentation rely on the
agentdbpackage, which is fetched and executed vianpx agentdb@latestfrom the NPM registry. It also utilizes theagentic-flowlibrary for API integrations. These resources are official components provided by the skill's author (ruvnet). - [COMMAND_EXECUTION]: The skill documentation provides several shell commands for database initialization (
agentdb init), plugin creation (agentdb create-plugin), and starting an MCP server (agentdb mcp). These commands are standard for managing the AgentDB lifecycle and integrating it with platforms like Claude Code. - [INDIRECT_PROMPT_INJECTION]: As a memory management system, the skill is designed to store and retrieve interaction data, including conversation history and user preferences. This creates a surface for indirect prompt injection where malicious instructions embedded in stored data could potentially influence the agent when the memory is retrieved. The skill uses
JSON.stringifyfor data handling, but does not explicitly demonstrate input sanitization or boundary markers in its code snippets. - Ingestion points: Data enters the system via
adapter.insertPattern,db.storeMemory, anddb.storeFactcalls inSKILL.md. - Boundary markers: None identified in the provided implementation snippets.
- Capability inventory: The skill facilitates file writing (database files), vector search, and model training operations as shown in
SKILL.mdand CLI examples. - Sanitization: The snippets show structured data handling (JSON), but no specific filtering for prompt injection patterns.
Audit Metadata