Skip to content
🧠 Research & Foundations phase — building the KB from the ground up. See the roadmap →

2026-04-11 · Reinit + full docs site build-out

CreatedUpdated

Reinitialized cyberbaser after ~4 months dormant. The working state was: a PARA-structured docs/ of implementation notes, a parked Phase-0 Astro prototype in site/, an almost-empty .claude/, and no real first-principles thinking anywhere.

The reinit had two phases:

Phase 1 — workspace structure (early in the session)

  • Consolidated site/ → docs/ to match the crosswalker / cyberchaste convention (both sibling projects use docs/ as the Astro Starlight root, not site/).
  • Archived the old PARA dev docs to .workspace/_archive-phase-0-docs/ (gitignored contents, preserved for mining).
  • Created .workspace/ as a tracked-but-ignored scratch folder via .gitignore rules.
  • Scaffolded .claude/ with PROJECT_CONTEXT, FOCUS, KNOWLEDGE_BASE_PHILOSOPHY (copied verbatim from crosswalker), DOCUMENTATION_STYLE, RESEARCH_SOURCES, and numbered first-principles files (00-INDEX through 41-QUESTIONS-RESOLVED).
  • Rewrote root CLAUDE.md to point at .claude/ as the first-read surface.

Phase 2 — docs site build-out (later in the session)

  • Full template copy from crosswalker: Astro 6, Starlight 0.38, Nova theme, full plugin stack (site-graph, blog, announcement, image-zoom, tags). Omitted starlight-heading-badges because its <starlight-toc> custom element collided with Nova’s at runtime.
  • Wired remark-obsidian-callout, remark-wiki-link, rehype-mermaid, rehype-external-links.
  • Custom components: PageTitle.astro (status + date meta) and MobileMenuFooter.astro, both rebadged to .cb-* classes.
  • Silver + emerald palette in brand.css. Graphite base (near-black in dark, off-white in light) with emerald accent appearing only in links, focus, CTAs, active sidebar, hero gradient, card top borders on hover.
  • Ported .claude/ numbered files into real Concepts / Design / Reference prose. Each custom MDX component uses hand-drawn inline SVG (no Mermaid walls): triangle tradeoff diagram on Problem, 4-layer architecture diagram on Architecture, flow diagram on the splash.

A bunch of MDX + Astro gotchas surfaced as the first screenshots came in:

  1. Rotating headline was a transparent blob. Nested background-clip: text on parent h1 + child rotate-wrap; the child inherited color: transparent and painted the box background. Fix: parent headline is solid silver; only the rotate wrap gets the gradient.
  2. Flow SVG was all black. Inline MDX <style>{}</style> wrapped rules in :global(...), which is an Astro-scoped-component escape hatch that doesn’t work inside .mdx <style> tags — rules were silently dropped. Fix: moved all .cb-sv-* rules into brand.css.
  3. Hero subhead had height: 0. MDX was splitting multiline <p class="cb-hero-subhead">...</p> into an empty classed <p> + an unclassed <p> with the actual text. Fix: single-line MDX for all custom JSX elements containing text.
  4. <h1><p>...</p></h1> on the hero headline. Same MDX wrapping bug. Fix: changed to <h2> (which also fixed duplicate-h1 semantics since Starlight’s PageTitle already renders the <h1>), collapsed to single line, updated PageTitle.astro to hide the default h1 on splash pages via sidebar.hidden === true.
  5. starlight-toc already defined runtime error. starlight-heading-badges registered a custom element that collided with Nova’s. Removed the plugin.
  • Added tests/screenshots.spec.ts that captures 30 full-page screenshots (15 pages × dark/light) for design iteration. Run serially with --workers=1 to avoid overwhelming the dev server.
  • Added scripts/preflight.mjs — cross-platform bun install guard so bun run dev works from both WSL and Windows PowerShell without manual reinstalls.
  • bc02aa7 — Reinitialize workspace: adopt .claude/ first-principles KB convention
  • 31bb90b — Build out docs site: template copy, silver+emerald theme, first-principles content

Phase R is still active. The exit criteria in Roadmap haven’t moved: Problem / Ecosystem / Concepts / Prior Art all have first-pass content but need deeper research, especially Prior Art where each tool entry should grow from “what I know” into “what I actually tested.” The Open Questions in Reference are the immediate blockers.