Encyclopedia · Vol. I–X

The field, operational.

Canon is a research encyclopedia of agentic and multi-agent engineering — and an operating layer for agents. 34 SKILL.md packs, job packs, playbooks, recipes, and a machine API so a coding agent loads what it needs instead of ingesting the atlas.

Open the skills Open the field

Skills: 34
Articles: 60
Repos admitted: 56
Volumes: 10

Volumes

Ten books. One loop. A field agents can run.

Vol. I

First Principles

What an agent is, and is not

Vol. II

Production Doctrine

The 12 factors and the outer loop

Vol. III

Single-Agent Patterns

Workflows that earn their extra hop

Vol. IV

Multi-Agent Architectures

Crews, graphs, swarms, and societies

Vol. V

The Harness

Skills, subagents, sandboxes, files

Vol. VI

Context & Memory

What the model is allowed to see

Vol. VII

Protocols & Tools

MCP, A2A, ACI, humans

Vol. VIII

Production Systems

Durable, observable, evaluated

Vol. IX

The Agent-Computer Interface

Permissions, injection, identity, the visible loop

Vol. X

Field Operations

How coding agents actually work

How a repo makes the cut

1. Distinctive contribution. A new topology, a protocol, a harness shape, or a production principle — not a wrapper around someone else’s README.
2. Review density. Stars as a floor (generally 10k+), plus forks, real issues, and recent maintenance. Viral 2026 star-farms with disabled PRs did not enter.
3. An agent can learn from it. Docs, papers, or source that encode a transferable practice.
4. Institutional or multi-year signal. Labs (Anthropic, OpenAI, Google, Microsoft, Hugging Face, Stanford, ByteDance) or independents with production evidence (LangGraph, CrewAI, Aider, Langfuse).