AgentDB Learning Plugins

This file is primarily documentation and usage examples for an AgentDB learning plugin system. There is no direct malicious code in the text itself, no hardcoded credentials, and no obvious obfuscated payload. However, it instructs users to run code via npx (download-and-execute), and the API examples call out to functions whose implementations (computeEmbedding, adapter.train, adapter.retrieveWithReasoning) may perform network activity, remote training, telemetry, or data syncing. Those runtime behaviors are not specified here, creating a supply-chain risk: installing/running remote packages and adapters can execute arbitrary code and possibly exfiltrate stored experiences/embeddings or forward credentials. Recommendation: treat this skill as medium-risk until the actual agentdb/agentic-flow package and adapter implementations are audited. Verify where computeEmbedding and adapter functions execute (local WASM vs hosted endpoints), avoid running unpinned npx installs in privileged environments, and review transitive dependencies for unexpected network calls or credential forwarding.