AgentDB Vector Search

Pass

Audited by Gen Agent Trust Hub on Mar 4, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill utilizes npx agentdb@latest to execute code from the NPM registry. As this is the vendor's own package and uses a well-known registry, it is considered a standard operational dependency.
  • [COMMAND_EXECUTION]: Provides numerous CLI instructions for database management, including initialization, querying, and stats retrieval. It also details integration with Claude Code via MCP.
  • [PROMPT_INJECTION]: The skill exhibits an Indirect Prompt Injection surface (Category 8) in its RAG implementation. * Ingestion points: The skill reads potentially untrusted data from vectors.db or backup.json and processes search results. * Boundary markers: The provided prompt template uses basic 'Context:' and 'Question:' delimiters which do not strongly isolate retrieved content from the agent's instructions. * Capability inventory: The skill has the ability to run shell commands, write to the file system, and access the OpenAI API for embeddings. * Sanitization: No explicit filtering or sanitization of retrieved database content is described before it is injected into the LLM prompt.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 4, 2026, 07:39 PM