A standard for websites that work as well for AI agents as they do for humans.
- Landing page: agentfirst.globalbr.ai
- Formal spec: agentfirst.globalbr.ai/spec
- Verifier: agentfirst.globalbr.ai/checker
- Reference implementation: WikiHub
A draft, living specification that organizes "agent-first" web capabilities around four orthogonal environments (Browser agents / Headless coding agents / MCP-client agents / In-site agents) instead of a single progressive conformance ladder, and around ten concrete patterns ordered by impact-per-hour-of-work.
The standard is driven by production experience against real implementations
(WikiHub, ListHub) and tries hard to be honest about what actually works in
the current agent ecosystem — including the parts that don't (WebMCP is
Chrome-only, llms.txt doesn't drive traffic, consumer-grade agent identity
doesn't exist yet).
| File | What |
|---|---|
overview.md |
Non-normative overview (same content as the landing page) |
spec.md |
Formal specification using RFC 2119 keywords |
checker/ |
Python source for the /check?url=… verifier |
LICENSE |
CC BY 4.0 |
Open an issue or pull request. Small fixes (typos, clarifications) go straight in. Substantive changes that alter the spec's shape (new environments, new required patterns, changes to badge criteria) should be proposed as an issue first so we can talk about tradeoffs in public.
Implementers who want to be listed as conforming examples in §7 of the spec can
open a PR to spec.md with a link and the environments they cover.
CC BY 4.0. Use it, fork it, credit the project.