feat(docs): catalog auto-generation script and LLM layer (Plan 10 Phase C)

[LastStep]

Summary

Add a catalog generation script that reads catalog/*/meta.yaml source files and auto-generates data tables in the docs site catalog pages, plus a machine-readable catalog.json. Configure the starlight-llms-txt plugin with curated Bonsai-specific content and 4 custom topic bundles.

Changes

• website/scripts/generate-catalog.mjs — Node.js script that reads all catalog meta.yaml/agent.yaml files, regenerates tables between MDX marker comments in 7 catalog pages, and outputs public/catalog.json
• website/package.json — Added generate:catalog npm script
• website/astro.config.mjs — Configured starlightLlmsTxt() with project name, description, details, and 4 custom sets (Concepts, Commands, Catalog, Configuration)
• website/public/catalog.json — Generated JSON dump of all 57 catalog items with full metadata
• website/src/content/docs/catalog/*.mdx — Added {/* CATALOG-TABLE-START/END */} markers around data tables; fixed data drift in skills.mdx (design-guide description/agents, review-checklist agents, test-strategy agents); sorted sensor table alphabetically

Plan

Plan 10, Phase C: station/Playbook/Plans/Active/10-docs-site.md

Verification

• npm run generate:catalog — runs cleanly, reports 57 items
• npm run build — builds without errors (45 pages, all links valid)
• Catalog tables match source meta.yaml (design-guide, review-checklist, test-strategy drift fixed)
• catalog.json contains all 57 items (6 agents + 17 skills + 10 workflows + 4 protocols + 12 sensors + 8 routines)
• llms.txt has curated Bonsai content (custom description, install instructions, agent types)
• Custom set bundles generated: _llms-txt/concepts.txt, commands.txt, catalog.txt, configuration.txt
• Script is idempotent — second run produces no changes

🤖 Generated with Claude Code