OKF publishing conventions for oriz bundles
OKF publishing conventions
Decision
Publish oriz OKF bundles with:
Required (Kiso defaults + Google spec)
llms.txtat root — machine-readable navigation for LLMs.sitemap.xmlat root — SEO + agent crawl.index.mdin every folder — progressive disclosure.log.mdin every folder with chronological changes.<slug>.mdper concept — OKF frontmatter compliant.
Additions (oriz first-mover)
feed.xml(RSS 2.0) at root — new/updated concept files last 30 days.atom.xmlat root — same content, Atom variant./schema.jsonat root — machine-readable OKF version + custom fields (confidence,durability)./related.jsonat root — precomputed graph edges for boone-style traversal without full re-index.
Optional (per-bundle)
/opensearch.xmlfor Firefox/Chrome address-bar search./manifest.jsonif the bundle acts as PWA-installable knowledge base.
Why RSS/Atom
Deep-research 2026-07-02 confirmed no OKF RSS/Atom convention exists in the ecosystem. First-mover:
- Agents can subscribe to knowledge updates.
- Humans use standard readers.
- Drives ecosystem convergence — if adopted, becomes de-facto standard.
Why /schema.json + /related.json
- schema.json — declares OKF version consumers should expect. Enables version-aware clients.
- related.json — cheap graph queries without needing full boone install client-side.
Publish path
knowledge.oriz.in/skills.oriz.in/- Both same convention. Both Kiso-generated + oriz-specific post-processors for RSS/Atom + schema/related.
Anti-patterns
- ❌ Non-standard filenames (e.g.
updates.mdinstead oflog.md) — breaks Kiso + agent expectations. - ❌ Ship without llms.txt — the whole point of OKF is LLM-friendly.
- ❌ Ship without validator gate — bad frontmatter propagates to public site.