Open numbered standards for the markdown files agents read, write, and run from.
Code freezes intent at compile time. Text re-interprets it on every run.
AIP = Agentproto Improvement Proposal. Same model as ERCs (Ethereum), BIPs (Bitcoin), PEPs (Python) — but for the markdown primitives every agent runtime is converging on (AGENTS.md, SKILL.md, CLAUDE.md, OPERATOR.md, KNOWLEDGE.md, …).
One numbered spec per primitive. One place to look it up. One shared vocabulary across runtimes.
Rendered docs: https://agentproto.sh/docs · Status: 0.1.0-alpha
| Repo | Contents | License |
|---|---|---|
agentproto |
Markdown AIP specs + canonical resources + RFCs | CC-BY-4.0 (specs) / MIT (samples) |
ts |
TypeScript reference runtime + framework adapters | MIT |
site |
Next.js renderer at agentproto.sh | MIT |
- AIP-1 — purpose & process
- AIP-9 —
OPERATOR.md, the shell that ties identity, memory, capabilities, governance - AIP-14 —
TOOL.md, the abstract tool contract every driver subtype specialises - AIP-7 —
GOVERNANCE.md, how approvals, audit logs, and autonomy policies are recorded as files
New AIPs are submitted as PRs against agentproto/agentproto — copy specs/aip-2.mdx as the template and follow CONTRIBUTING.md. Community expectations live in CODE_OF_CONDUCT.md.