Keeping agent readiness from regressing
Readiness is a property of this week's release, not a project you finish.
In short
Agent readiness breaks quietly: a redesign moves pricing into a client-side widget, a new bot rule blocks an assistant, a login change adds a challenge. Put file checks in CI, re-run key tasks after releases that touch pricing, sign-in, bot rules or checkout, keep a dated log, and watch the standards and scanner methods for changes.
What usually breaks
- Rendering: a page rebuilt as a client-side app returns an empty shell to agents that do not run JavaScript.
- Access: a new CDN or firewall rule blocks a crawler or agent that robots.txt allows.
- Links: llms.txt still points to pages that moved.
- Forms: a redesigned form loses its labels or adds a custom control.
- Challenges: a bot challenge appears on a step an agent used to pass.
Each one fails a layer from lesson 1, so the fix is known. The hard part is noticing.
Catch breaks before agents do
Run file checks on every deploy. A CLI makes this practical: Ora's npx ax audit <domain> scores a URL from a terminal or CI job, and its MCP server can rescan after each fix. Scripted requests work too. Fetch the pricing page and assert that the plan names are in the HTML. Fetch /llms.txt and assert that every link returns 200. Request a key page with Accept: text/markdown and check the content type.
When to re-run task tests
Re-run them after any release that touches pricing, sign-in, bot rules or checkout, and on a fixed schedule otherwise. Keep the same tasks and the same expected answers so runs can be compared, and date every entry in the log.
How standards changes affect you
The standards moved several times in 2026. A2A reached v1.0.0 on 12 March, the MCP specification of 28 July made the protocol stateless, and llms.txt was revised to v2 in August. AgentReady v2.0 followed in October and retired AR-READ-10. A change can make a working setup outdated, such as an MCP server that relies on sessions. It can also add something worth adopting, such as llms.txt link relations. The news page lists these releases with dates and sources.
Telling a site change from a method change
When a score moves, check two things: what you shipped since the last scan, and whether the scanner changed its checks. Method changes are common and dated:
- Ora's catalog stood at 132 checks under contract 1.38.0 on 4 November 2026, while its methodology page still stated 126.
- agent-ready.dev stopped crediting catch-all pages as llms.txt on 20 October 2026 and did not re-score historical scans.
- Cloudflare's scanner launched with four dimensions in April 2026 and listed 23 checks in five categories by 4 November 2026.
Record the date and the method with each score, and the cause is easy to find.
Sources
- Ora, home page. https://ora.ai/, read 2026-11-04.
- Ora, check catalog (contract 1.38.0). https://ora.ai/api/checks, read 2026-11-04.
- Ora, methodology. https://ora.ai/methodology, read 2026-11-04.
- agent-ready.dev, methodology. https://agent-ready.dev/methodology, read 2026-11-04.
- Cloudflare, isitagentready.com llms.txt. https://isitagentready.com/llms.txt, read 2026-11-04.
- AgentReady changelog. https://github.com/agentready-org/standard/blob/main/CHANGELOG.md, read 2026-11-04.
- A2A releases. https://github.com/a2aproject/A2A/releases, read 2026-09.
- MCP blog, the 2026-07-28 specification. https://blog.modelcontextprotocol.io/posts/2026-07-28/, read 2026-09.
- llms.txt changes. https://llmstxt.org/changes.html, read 2026-09.
Last reviewed 2026-11-05