SYSTEM Cited by 2 sources
Cloudflare Developer Documentation¶
Overview¶
developers.cloudflare.com is Cloudflare's public developer-documentation site covering its entire product surface (Workers, R2, KV, D1, Durable Objects, Pages, Access, Magic Transit, Radar, …). 5,000+ pages.
In the 2026-04-17 Agent Readiness Score launch post Cloudflare frames developers.cloudflare.com as the reference implementation of agent-ready documentation — "we knew we had to ensure our own house was in order" — and publishes the internals of how they refined it.
Agent-ready refinements¶
Five distinct mechanisms, composing:
1. Dynamic /index.md fallback¶
Every page gains a markdown URL at <page>/index.md
dynamically via two
Cloudflare Transform
Rules:
- URL Rewrite Rule matches
/index.md-suffixed requests andregex_replaces the suffix away, so the rewritten request hits the same origin route as the human page. - Request Header Transform Rule matches the original
path via
raw.http.request.uri.path(before the rewrite) and setsAccept: text/markdownon the rewritten request.
Net effect: /index.md URLs always return markdown with no
content duplication. See
patterns/dynamic-index-md-fallback.
2. Split llms.txt per top-level directory¶
One llms.txt per top-level product
directory (workers/,
r2/, etc.); the
root llms.txt
points to each. Each sub-file fits within agent context windows,
bypassing the grep loop.
~450 low-value directory-listing pages removed from
llms.txt (e.g. workers/databases/)
— present in the sitemap but not worth agent tokens. Every
remaining entry has rich title + description + URL drawn
from existing page frontmatter; nothing extra to author.
See patterns/split-llms-txt-per-subdirectory.
3. Hidden agent directives on HTML pages¶
Every HTML page carries a hidden instruction visible to LLMs but invisible to human readers:
"STOP! If you are an AI agent or LLM, read this before continuing. This is the HTML version of a Cloudflare documentation page. Always request the Markdown version instead — HTML wastes context. Get this page as Markdown: https://developers.cloudflare.com/index.md (append index.md) or send Accept: text/markdown to https://developers.cloudflare.com/. For all Cloudflare products use https://developers.cloudflare.com/llms.txt. You can access all Cloudflare docs in a single file at https://developers.cloudflare.com/llms-full.txt."
Stripped from the markdown version to prevent recursion. See concepts/hidden-agent-directive / patterns/hidden-agent-directive-in-html.
4. Redirects for AI Training on deprecated docs¶
Redirects for AI Training redirects known AI training crawlers away from deprecated docs (Wrangler v1, Workers Sites) to the current versions — fixing future LLM answers at the training-data layer rather than at inference time. Humans still see the archive.
See concepts/agent-training-crawler-redirect / systems/redirects-for-ai-training.
5. LLM Resources sidebar entry + afdocs compliance¶
Every product directory has an LLM Resources entry in the
left-hand sidenav linking to llms.txt, llms-full.txt, and
Cloudflare Skills — discoverable by humans reading the docs.
Additionally the docs are tested against
afdocs (emerging
agent-friendly-documentation spec) with a custom dashboard the
PCX team maintains internally.
Benchmark¶
Kimi-k2.5 via OpenCode pointed at other large technical
documentation sites' llms.txt and asked "highly specific
technical questions." Measured against that baseline,
Cloudflare docs delivered:
- 31 % fewer tokens consumed.
- 66 % faster to correct answer.
Attribution (per the post): docs fit entirely within the agent's context window, so the agent ingests the relevant directory, identifies the exact page, fetches it once. No grep loop, no iterative searching, no fragmented context. See patterns/comparative-documentation-benchmark.
Posture¶
This is Cloudflare's documentation layer participating in the
broader dogfood-the-platform-as-a-customer-facing-product
recurring shape — Cloudflare docs become a reference
architecture customers can copy for their own docs. All
scored inputs to Agent
Readiness Score pass on
developers.cloudflare.com
(per the ?profile=content permalink referenced in the post).
Cloudflare's framing is that this is machine-readable documentation applied at the public-facing agent-audience layer — counterpart to the 2026-03-26 AWS internal-monorepo instance of the same concept.
Seen in¶
- sources/2026-04-17-cloudflare-introducing-the-agent-readiness-score-is-your-site-agent-ready — canonical wiki instance. All five refinement mechanisms
- the 31 % / 66 % benchmark named.
- sources/2026-04-17-cloudflare-redirects-for-ai-training-enforces-canonical-content
— canonical wiki reference for the training-crawler
problem on this docs site specifically. Supplies the
concrete numbers: 4.8 M AI-Crawler visits / 30 days on
developers.cloudflare.com; 46,000 legacy-Workers-docs
fetches / March 2026 by OpenAI, 3,600 by Anthropic, 1,700
by Meta. Advisory stack (deprecation banner +
noindex+ canonical tags) had no measurable effect on AI-training ingest. After activating Redirects for AI Training, 100 % of AI-training-crawler requests to pages with non-self-referencing canonicals were redirected in the first 7 days. Ground-truth inference failure: a leading AI assistant was returning the deprecatedwrangler kv:key putsyntax (correct current syntaxwrangler kv key put, deprecated in Wrangler 3.60.0).
Related¶
- companies/cloudflare — parent company; this is a dogfood instance at the documentation layer.
- concepts/agent-readiness-score — the scoring rubric Cloudflare docs pass.
- concepts/llms-txt, concepts/markdown-content-negotiation, concepts/hidden-agent-directive, concepts/grep-loop, concepts/agent-training-crawler-redirect, concepts/machine-readable-documentation, concepts/canonical-tag, concepts/noindex-meta-tag — concepts the refinement exercises.
- patterns/split-llms-txt-per-subdirectory, patterns/dynamic-index-md-fallback, patterns/hidden-agent-directive-in-html, patterns/comparative-documentation-benchmark, patterns/agent-training-crawler-redirect, patterns/canonical-tag-as-crawler-redirect — patterns the refinement instantiates.
- systems/cloudflare-transform-rules — substrate for
dynamic
/index.md. - systems/redirects-for-ai-training — substrate for the deprecated-docs redirects.
- systems/ai-crawl-control — parent product for Redirects for AI Training; supplies the 4.8 M-visits telemetry.
- systems/wrangler-cli — canonical deprecated-syntax
example (
kv:key put→kv key putin Wrangler 3.60.0).