# Changelog All notable changes to the AgentReady spec are recorded here. The readable surface is [agentready.org](https://agentready.org); the text lives in [`spec/spec.md`](spec/spec.md). --- ## v2.0 — October 2026 A major version: `AR-ACT-10` adds a `MUST`, so a product that conformed to v1.0 may not conform to v2.0, and `AR-ACT-07` WebMCP is raised from `MAY` to `SHOULD`. - New practice: `AR-ACT-10` an MCP server or a CLI (MUST, at least one) for products agents are expected to act on. MUST now marks what an agent fails without: the two changes that broke an agent in the studies, and an interface to act through. `AR-ACT-03` (MCP server) and `AR-ACT-06` (SDKs and a CLI) stay SHOULD as the specific forms. - `AR-ACT-07` WebMCP (SHOULD): register tools on `document.modelContext` when the product has a web UI an agent should operate. Previously on agentready.org without an identifier, as a MAY. - New practice: `AR-FIND-04` AI Catalog at `/.well-known/ai-catalog.json`, with Agentic Resource Discovery (ARD) as the registry layer (MAY). - New practice: `AR-READ-11` pricing on a first-party page (SHOULD), backed by pricing answers being 60% accurate from the business's own pages against 37% from elsewhere. It takes the next unused number: `AR-READ-10` was removed in August 2026 and stays retired. - New practice: `AR-ACT-08` agent payments (MAY): x402 or MPP for per-call pricing, UCP or ACP for checkout, and Stripe Link (Link on checkout and Shared Payment Tokens) for agents paying with a Link Agent Wallet, backed by Ora's October 2026 checkout study (92.3% of Link test checkouts done right). - New practice: `AR-ACT-09` NLWeb `/ask` endpoint with an `ask` tool over MCP (MAY). - `AR-FIND-01` now covers bot protection as well as `robots.txt`: let verified agents through the WAF or bot manager, and throttle with a `429` instead of a challenge page. - The find section opens with what happens when agents cannot read a site, from Ora's AX is the New AEO study (1,056 business sites): the agent answers anyway, and those answers are 3.7× as likely to miss every fact asked for. - `[STUDIES]` is described by the full research base behind the spec: more than 250,000 real agent runs, the controlled experiment, and the Ora research reports cited next to their numbers. The per-claim numbers keep their raw counts from the published 1,033-run subset in [`data/traces.csv`](data/traces.csv). - The crawl-policy notes name agent search APIs (Brave Search API, Exa, Parallel, Tavily) as a crawler kind the wildcard rule covers. - The `AR-ACT-*` category applies to products agents are expected to act on, not only read about. ### Removed - `AR-READ-10` (`AGENTS.md` / `SKILLS.md`, MAY), removed in August 2026 along with its open question. The identifier is retired. --- ## v1.0 — August 2026 First public version of **"Make your site readable by AI agents."** The spec was restructured from five topic sections (Discoverability, Content, Capabilities, Identity & Access, Commerce) into the order in which agents encounter a site — find, read, act — with every requirement grounded in the studies. Stable identifiers moved from `AR-DISC-*` / `AR-CONT-*` / `AR-CAPA-*` / `AR-IDEN-*` / `AR-COMM-*` to `AR-FIND-*` / `AR-READ-*` / `AR-ACT-*`; the old identifiers are retired, not reused. Strengths (`MUST` / `SHOULD` / `MAY`) are now assigned from measurement: the two changes that broke an agent in the studies are `MUST`, practices that measurably shaped where agents went are `SHOULD`, emerging conventions are `MAY`. Figures are tagged by source (`[STUDIES]` for traced agent runs and the controlled experiment, `[ORA SCANNER]` for reachability checks). Removed in this restructure: - The `/spec.json` machine-readable artifact (the markdown source in `spec/spec.md` is the machine-readable form for now). - The baseline vs. conditional applicability model; applicability is stated in each requirement's text. The v1.0.0 entry below is the version announced on March 20, 2026 ([ora.ai/blog/agentready-open-standard](https://ora.ai/blog/agentready-open-standard)) and published on April 24, 2026. --- ## v1.0.0 — 2026-04-24 First published version of AgentReady. ### Added - Five sections, organized by topic rather than by agent journey: - **Discoverability** (`AR-DISC-01` … `AR-DISC-06`) - **Content for agents** (`AR-CONT-01` … `AR-CONT-04`) - **Capabilities** (`AR-CAPA-01` … `AR-CAPA-09`) - **Identity & Access** (`AR-IDEN-01` … `AR-IDEN-06`) - **Commerce** (`AR-COMM-01` … `AR-COMM-05`) - Normative strength (`MUST` / `SHOULD` / `MAY` per BCP 14) on every requirement. - Applicability model (`baseline` vs. conditional) on every requirement. - Stable identifiers for every requirement, preserved across revisions. - Conformance definition, including the treatment of Commerce as a set of alternative, interoperating protocols where at least one applies. - Machine-readable artifact at `/spec.json`. - Agent-readable summary at [`/llms.txt`](https://agentready.org/llms.txt). ### Requirements Discoverability - `AR-DISC-01` robots.txt with AI policy — **MUST** (baseline) - `AR-DISC-02` Sitemap — **SHOULD** (baseline) - `AR-DISC-03` llms.txt — **SHOULD** (baseline) - `AR-DISC-04` llms-full.txt — **MAY** (baseline) - `AR-DISC-05` HTTP Link response header — **MAY** (baseline) - `AR-DISC-06` NLWeb schema feeds — **MAY** (conditional) Content for agents - `AR-CONT-01` JSON-LD structured data — **SHOULD** (baseline) - `AR-CONT-02` Markdown content negotiation — **MAY** (baseline) - `AR-CONT-03` /index.md fallback — **MAY** (baseline) - `AR-CONT-04` Speakable content markup — **MAY** (conditional) Capabilities - `AR-CAPA-01` Model Context Protocol (MCP) — **MUST** (conditional) - `AR-CAPA-04` A2A Agent Card — **MUST** (conditional) - `AR-CAPA-08` OpenAPI — **MUST** (conditional) - `AR-CAPA-02` MCP Server Card — **SHOULD** (conditional) - `AR-CAPA-03` MCP Apps — **MAY** (conditional) - `AR-CAPA-05` WebMCP — **MAY** (conditional) - `AR-CAPA-06` NLWeb /ask endpoint — **MAY** (conditional) - `AR-CAPA-07` Agent Skills — **MAY** (conditional) - `AR-CAPA-09` ai-plugin.json — **MAY** (conditional) Identity & Access - `AR-IDEN-02` OAuth 2.0 — **MUST** (conditional) - `AR-IDEN-03` OAuth Authorization Server Metadata — **MUST** (conditional) - `AR-IDEN-05` PKCE — **MUST** (conditional) - `AR-IDEN-01` Web Bot Auth — **SHOULD** (baseline) - `AR-IDEN-04` OAuth Protected Resource — **SHOULD** (conditional) - `AR-IDEN-06` API Catalog — **MAY** (conditional) Commerce - `AR-COMM-01` x402 — **MAY** (conditional) - `AR-COMM-02` Agentic Commerce Protocol (ACP) — **MAY** (conditional) - `AR-COMM-03` ACP Delegate Payment — **MAY** (conditional) - `AR-COMM-04` Universal Commerce Protocol (UCP) — **MAY** (conditional) - `AR-COMM-05` Machine Payments Protocol (MPP) — **MAY** (conditional) All v1.0.0 identifiers were retired in the v1.0 restructure.