Skip to content

CLAUDE.md

Latest commit

 

History

History
31 lines (23 loc) · 2.42 KB

CLAUDE.md

File metadata and controls

31 lines (23 loc) · 2.42 KB
Raw

Design System

Always read DESIGN.md before making any visual or UI decisions. All font choices, colors, spacing, layout, and aesthetic direction are defined there. Do not deviate without explicit user approval.

Key rules at a glance:

  • Public web is content-first: show real Sessions and authors before feature explanations
  • Share explicitly sends records; supported Claude/Codex Shares are Public by default, while unsupported providers remain Link-only
  • Public metadata is author-attributed (@handle · published 2h ago), never first-person
  • Paperboy electric blue accent #1387FF (light) / #5BB1F0 (dark) is the sole product accent; amber is reserved for warning semantics
  • Void palette: #000000 background with #090909 surfaces in dark mode; white with neutral-gray surfaces in light mode
  • Geist Sans for UI chrome; Geist Mono for Session records, commands, URLs, and paths
  • Session pages separate interpretive Summary from machine-derived evidence
  • Visibility and continuation lineage must remain explicit trust signals
  • Team · {name} is a real tenant boundary: current members only, private, no-store, and never included in public Profile, Explore, search, RSS, previews, or OG metadata
  • Emoji are placeholder icons only — production UI uses Lucide React SVGs
  • "via ACP · local" remains mandatory wherever local AI synthesis appears

In QA mode, flag any code that doesn't match DESIGN.md.

Test discipline

Every bug fix and feature PR must:

  1. Add tests for the change. Bug fix → a regression test that fails on the pre-fix code. Feature → primary path + non-obvious edges (empty / error / boundary). Web UI changes use focused component and route tests under apps/web; pure logic uses vitest under {apps,packages}/*/src/**.test.ts.
  2. Run the adjacent suite, not just the new tests, before declaring done. Changes ripple: virtualization breaks DOM-count assertions, selector renames break old e2e, fixture changes shift sort order. Fix any cascading failures in the same PR — never ship a regression with a TODO.
  3. Don't fight flakiness. A flake is a test that's lying. Diagnose root cause once; if it can't be made reliable without fighting the framework, drop it and document the coverage gap in the PR body rather than papering over with --repeat-each.

Completion checklist: typecheck clean → new tests green → adjacent suite green → flaky candidates stress-run 2–3× → only then declare done.