Skip to content

add(considered): AGENTS.md is a repository convention, not a website one - #213

Draft
jdevalk wants to merge 1 commit into
mainfrom
considered/agents-md-2026-09-12
Draft

add(considered): AGENTS.md is a repository convention, not a website one#213
jdevalk wants to merge 1 commit into
mainfrom
considered/agents-md-2026-09-12

Conversation

@jdevalk

@jdevalk jdevalk commented Sep 12, 2026

Copy link
Copy Markdown
Owner

What changed

Adds src/content/considered/agents-md.md — one entry in the public /considered/ register. No spec page, no changelog entry, no derived surfaces touched.

Why now

Today's standards scan picked it up through the Agentic AI Foundation thread: A2A moved into AAIF on 17 August 2026, and AAIF's anchoring contributions are MCP, goose and AGENTS.md. The spec covers MCP and A2A and says nothing at all about AGENTS.md, which is now the odd silence — this is a spec about agent-readiness, and AGENTS.md is the single most widely adopted agent-facing Markdown convention there is.

It comfortably clears the adoption bar (60,000+ repositories, 20+ coding tools). Turning it down on adoption grounds would be wrong; turning it down silently would leave readers guessing. So: an entry, with the actual reason.

The reason — out-of-scope

Nobody serves it. Every agent-readiness topic in this spec describes something an origin answers over HTTP to an agent that arrived from outside and knows nothing about the site — llms.txt, Agent Skills discovery, A2A agent cards, Markdown source endpoints. AGENTS.md runs the other way: instructions that travel with the source, read by an agent that already holds a working copy. There is no URL to check, and a website is neither better nor worse for the file existing.

The entry also names Agent Skills discovery as the website-side equivalent, because the two are the same instinct pointed in opposite directions and are easy to confuse.

revisit: a served form of the convention — agents.md describing a path an origin answers, or consumers fetching AGENTS.md over HTTP from a site they never cloned. Direction is the open question, not adoption.

Sources

Verification caveat: the scan sandbox blocked direct fetches to agents.md and linuxfoundation.org, so these three URLs were confirmed via search-index results (title + URL returned live) rather than a direct GET. Worth a click before merge.

Checks

  • npm run build passes (170 pages, Pagefind index clean).
  • npm run format:check clean; pre-commit hook (eslint, prettier, skill-drift) passed — SKILL.md still agrees at 170 pages / 10 categories.
  • No spec page count change, so no OG regeneration and no sign:skill needed.

🤖 Generated with Claude Code

AGENTS.md passes the adoption bar easily — 60k+ repositories, 20+ tools,
now stewarded by the Agentic AI Foundation alongside MCP — so its absence
from an agent-readiness spec needs an explanation rather than silence.

It fails the direction test instead. Every agent-readiness topic here
describes something an origin answers over HTTP to an agent that arrived
from outside; AGENTS.md travels with the source to an agent that already
cloned it. No URL to check, no visitor or crawler outcome. The entry names
Agent Skills discovery as the website-side equivalent, since the two are
easy to confuse.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
@cloudflare-workers-and-pages

Copy link
Copy Markdown

Deploying specification-website with  Cloudflare Pages  Cloudflare Pages

Latest commit: 4e7440d
Status: ✅  Deploy successful!
Preview URL: https://24d4c456.specification-website.pages.dev
Branch Preview URL: https://considered-agents-md-2026-09.specification-website.pages.dev

View logs

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant