ReasoningBank with AgentDB

Pass

Audited by Gen Agent Trust Hub on Feb 26, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill utilizes npx agentdb@latest which downloads and executes the AgentDB CLI directly from the npm registry. This is considered normal vendor functionality for this skill.
  • [COMMAND_EXECUTION]: CLI commands are used to initialize the database and set up a Model Context Protocol (MCP) server via npx agentdb@latest mcp, which allows the agent to interface with the local vector database.
  • [PROMPT_INJECTION]: The skill implements a learning system that persists agent trajectories, creating a potential surface for indirect prompt injection.
  • Ingestion points: Task outcomes and step-by-step trajectories are ingested via the rb.insertPattern API and CLI migration/import tools.
  • Boundary markers: There are no explicit delimiters or 'ignore instructions' warnings shown in the code snippets to separate historical pattern data from active reasoning instructions.
  • Capability inventory: The skill provides a synthesizeContext feature which generates natural language narratives from past memories to guide agent behavior.
  • Sanitization: No evidence of content validation or sanitization is present for the ingested trajectory data, which could allow malicious patterns to influence future model outputs.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 26, 2026, 08:44 AM