v1.0.0 — the map of the modern agent stack

The build-your-own-x for the modern AI-agent stack: 10 components, each with the strongest from-scratch tutorial that exists — plus original, runnable reference implementations where none does.

Curated index (fetch-verified, re-checked weekly by CI):

• §1 agent loop · §2 tool calling · §3 memory · §4 RAG · §5 MCP · §6 coding agent · §7 token accounting · §8 evals · §9 multi-agent · §10 guardrails
• Editorial rule: no resource appears twice across the index.

Reference implementations — zero dependencies, no API key, Node 18+ (run in CI on Node 18/20/22):

• §5 — MCP server + client over stdio JSON-RPC: node reference/05-mcp-server-client/mcp-client.mjs
• §6 — coding agent: the loop + file/shell tools with real workspace containment: node reference/06-coding-agent/example.mjs
• §7 — token accounting: cost, prompt-cache math, streaming metrics: node reference/07-token-accounting/example.mjs

Trust layer:

• All three demos run on every push; every curated link is checked weekly (see badges).
• Contributions are gated by CONTRIBUTING criteria and a structured issue form.