feat: make PL R&D usable as AI context

[lksbrssr]

What this enables

Give an agent https://www.plrd.org/llms.txt, ask it to fetch focused resources, or download readable context to attach to a prompt. No chatbot, account, model dependency or special integration is needed for this public content.

• /ai/: human guide with copyable starter prompt, download, focused-area links and query examples.
• /llms.txt and downloadable /llms-full.txt.
• Topic/record Markdown, a schema-versioned /ai/index.json, bounded read-only /api/ai/search/ and direct JSON details.
• Footer + HTML discovery links. Existing shared content pipeline, not a second manually maintained catalog.

Coverage and safety

261 source-linked records: four area descriptors, 81 author profiles, 12 blog records, 79 publications, 82 talks and three top-level tutorials. Native blog/tutorial text, publication abstracts/metadata and talk summaries/recording links—not full papers or invented transcripts. Exact author display names and mixed-case source IDs are retained; three dangling talk bylines were corrected to their existing source display names.

This is explicitly a build-time snapshot, not a live CMS or whole-site dump. Live overview copy, ATProto posts, dashboards and nested tutorial lessons stay outside coverage. Canonical links point to current originals. No new reuse rights or guaranteed AI discovery are claimed.

Unlisted/draft/noindex/preview/private/future records fail closed in discovery and known-ID detail. Unpublished Impact/DHR/Open Lab and internal/account/edit surfaces are not exposed. Unknown raw fields are not serialized. Ordinary public API/preview/edit prose remains readable while private route references are rejected. No arbitrary URL fetch, source-code export, mutation endpoint, credentialed CORS or dependency change.

Verification

• pnpm 10 frozen install, 128 default tests, typecheck and production build passed.
• 818 actual local HTTP checks passed: every record's JSON, Markdown and source page; all topics; query success/errors; expected 404/405 boundaries.
• Separate parent source-parity and entrypoint checks passed. All 261 eligible source IDs match served results.
• Real Chrome: 1440/390/320px, clipboard, footer navigation and actual context download (254,802 bytes).
• Independent exact-final review PASS at 997b7d7c9b181c3dc6a3cb8a04b8849316511656. Located metadata-reference and slash/host false-positive findings were reproduced, corrected and independently closed. Final-SHA production build and all 818 served checks were then verified by the parent.
• Detailed local evidence and limits.

The build still reports external-data path warnings on unchanged pages (indexer query, fallback, oversized cache payloads), but exits 0. This feature does not claim to repair the existing live data system. GitHub also reports existing default-branch dependency alerts, including critical/high findings; this PR leaves dependency declarations and lockfiles unchanged. No merge or production deployment performed.

Preview

Use with AI preview — Vercel READY at the exact final SHA. Hosted entrypoints and source/query parity passed the same 13 + 18 parent checks. Guide prompt/index links deliberately use the future production URLs; this preview is not a claim that www.plrd.org/llms.txt is live yet.

Screenshots

Actual local Chrome captures of /ai/ at runtime SHA 0651f818dfc2f69b97859095b038f90c68870730; subsequent filtering hardening leaves the guide/component source unchanged and the complete current-corpus export byte-identical. No auth bypass or production configuration change.

Desktop

Mobile

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
plrd.org	Ready	Preview	Sep 11, 2026 1:21pm UTC