WP Agent Support Layer

WordPress plugin that makes a site discoverable, readable and operable by AI agents and crawlers ("agent-readiness"), without touching your infrastructure.

Requires WordPress 7.0+ and PHP 7.4+. Tested on PHP 7.4, 8.0, 8.2 and 8.3. No external requests, no telemetry.

Install

Download the latest zip from Releases and upload it from Plugins → Add New → Upload, or once published, install it from WordPress.org.

A git checkout is not installable as is: the HTML-to-Markdown library must be prefixed first. From the plugin directory run composer run build.

Configure

Tools → Agent Support Layer

• General: content types (default: posts and pages), regeneration interval (default: daily), items per run, generation status, Regenerate now.
• Signals: search, ai-input, ai-train and the experimental Content-Usage header.
• Crawlers: allow / block per AI crawler, preview of the generated robots.txt.
• llms.txt: description, introduction, items per section, llms-full.txt toggle and size limit.
• Manifests: toggle and contact email.
• Diagnostics: run the crawler simulation and review the report.

Each post has an Exclude from the agent layer checkbox in the editor sidebar.

Caches and CDNs

Content negotiation depends on Vary: Accept. Many page caches and CDNs ignore it and serve cached HTML to agents that ask for Markdown.

• Prefer the .md URLs (/my-post.md). They are separate resources with their own cache entries and their own Cache-Control: public, max-age=<regeneration interval>.
• If you cache HTML at the edge, either honour Vary: Accept or exclude requests whose Accept header contains text/markdown from the HTML cache.
• Do not strip response headers: Content-Signal, Content-Usage, X-Robots-Tag and Link are part of the feature.
• robots.txt, llms.txt, agent-skills.json and /.well-known/api-catalog may be cached for their Cache-Control lifetime.
• Cloudflare "Markdown for Agents" converts HTML at the edge. Keep only one converter: disable it to use the plugin's documents (front matter, llms.txt links), or keep it and use the plugin for signals, robots rules, llms.txt and manifests. Diagnostics warns when both are active.

Web server notes

Generated documents live in wp-content/uploads/wp-agent-support-layer/<token>/ and are only served through WordPress, which re-checks that the content is still public on every request.

• Apache: the plugin writes an .htaccess that denies direct access.

• nginx (does not read .htaccess):

  location ^~ /wp-content/uploads/wp-agent-support-layer/ { deny all; }

• .md URLs are resolved by the plugin at parse_request; no rewrite rules are needed beyond WordPress' own front controller.

• If a physical robots.txt or llms.txt exists in the site root, the web server serves it and the plugin shows what it would have generated.

WP-CLI

wp wpasl status                      # counters, last / next run, schedule
wp wpasl generate                    # run the current cycle to completion
wp wpasl generate --all              # regenerate everything
wp wpasl generate --post-type=page   # one type only
wp wpasl generate --batch            # a single scheduled-size batch
wp wpasl clear                       # delete generated files and reset state

If DISABLE_WP_CRON is set, call wp-cron.php from a system cron or run wp wpasl generate periodically.

Hooks for developers

The full list, with the file that documents each one, is in the "Filters and actions" section of readme.txt.

Development

composer install                # dev dependencies (PHPUnit, WPCS, polyfills)
composer run strauss            # prefix league/html-to-markdown into vendor-prefixed/
composer run install-wp-tests   # needs MySQL on 127.0.0.1:3306 (root/root)
composer test                   # PHPUnit against the official WordPress test suite
composer lint                   # PHPCS with WordPress Coding Standards + PHPCompatibility 7.4+

Specifications are written with OpenSpec in openspec/ before any code; the scenario-to-test map is in docs/spec-coverage.md. CI runs PHPUnit on PHP 7.4 / 8.0 / 8.2 / 8.3 (single site and multisite), PHPCS and the official Plugin Check on every push. Tagging vX.Y.Z builds the distributable zip and attaches it to a GitHub release.

Author and license

Developed by Mao Rodriguez (mao@111x.co). Supported by 111X S.A.S (111x.co, contacto@111x.co).

Licensed under the GPL-2.0-or-later. You may use, modify and redistribute this plugin under the terms of that license, which requires keeping the copyright and authorship notices intact.

En español

Plugin de WordPress que prepara un sitio para agentes y crawlers de IA: entrega en Markdown por Accept: text/markdown y URLs .md, Content Signals en robots.txt y cabeceras, reglas por crawler de IA, llms.txt, manifiestos de agente (agent-skills.json, OpenAPI 3.1, /.well-known/api-catalog) y un panel de diagnóstico de solo lectura. La interfaz está en inglés e incluye traducción al español.

Los documentos Markdown se generan por programación (WP-Cron) en lotes acotados y se guardan en uploads/wp-agent-support-layer/; guardar una entrada nunca dispara conversiones. Si usas nginx, añade la regla deny all indicada arriba; si usas una caché de página o CDN, apóyate en las URLs .md o haz que respete Vary: Accept. Las especificaciones se redactan con OpenSpec en openspec/ antes de escribir código.