The registry is a static pipeline: structured metadata in → typed pages out. Nothing is fetched at runtime; everything on the site is generated at build time from a single source of truth.
Pipeline stages
- Source records. Each resource is authored as a metadata record (slug, name, type, description, source, tags, plus category/stage/difficulty/verification fields).
- Generated data. A build-time script normalizes records, derives the taxonomy tree, stage mappings, filter facets, and stats (
stats.json), and emits JSON consumed by the pages. - Astro pages. Listing, detail, taxonomy, roadmap, and search pages are statically rendered from the generated data. Search runs client-side against the same index — there is no backend.
Resource → category → stage
Three independent classifications let you slice the same catalog three ways:
- Type answers what form it takes (a skill file, an MCP server, a runbook).
- Category answers what domain it belongs to (dimensional modeling, streaming, governance).
- Stage answers when in the lifecycle it applies (ingestion, quality, observability).
A single resource can address one stage while belonging to a category that spans many — the roadmap view joins resources to lifecycle stages.
Provenance and verification
Every record carries provenance fields: upstream source, license, and security notes where applicable. Verification status is assigned per resource; see Verification for the three levels. The date of the last check is stored as last_verified and is displayed on cards and detail pages.
Agent integration
Adoption follows a simple flow: pick a resource in the marketplace, read its detail page for configuration and any prerequisites, then install the artifact (skill file, MCP config, plugin) into your agent environment. Detail pages surface copy-ready configuration snippets where the resource format supports them.