AgentDB Learning Plugins

Pass

Audited by Gen Agent Trust Hub on Aug 21, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill uses npx agentdb@latest which fetches code from the NPM registry. It also references the source code hosted on GitHub and a vendor website.- [REMOTE_CODE_EXECUTION]: The instructions command the use of npx, which executes remote packages from NPM at runtime. This is standard behavior for the AgentDB CLI tools.- [COMMAND_EXECUTION]: The documentation provides multiple shell commands (create-plugin, list-templates, list-plugins) to be executed via the agentdb CLI.- [PROMPT_INJECTION]: The skill processes experience data patterns that are retrieved and used to generate context for agent decisions. This presents a potential surface for indirect prompt injection if the ingested data is sourced from an untrusted environment.
  • Ingestion points: Data enters the system via adapter.insertPattern in the experience collection logic.
  • Boundary markers: The provided code examples do not include explicit delimiters or instructions to ignore embedded commands within the experience data.
  • Capability inventory: The skill allows for local file writes (.agentdb/learning.db), command execution through npx, and ML model training.
  • Sanitization: There is no evidence of data sanitization or validation in the illustrative code snippets.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 21, 2026, 05:03 AM
Security Audit — agent-trust-hub — AgentDB Learning Plugins