When to read this: As a lookup. Catalogs every skill, slash command, and subagent the kit ships, with one-line purpose, invocation syntax, file location, and the chapter that explains it in depth.
Aesthetic commitment is mandatory before code. The phase produces: aesthetic name, 3 IS / 3 IS NOT adjectives, dominant color move, type contrast, motion temperament.
Phase 8 invokes design-reviewer automatically. Iterate diffs until PASS.
Scaffold a UI primitive with full state coverage (hover, focus-visible, active, disabled, loading, error, empty), story file, and variant matrix README. Auto-runs check-tokens.sh on exit.
Skills live at .claude/skills/<name>/SKILL.md after bootstrap. Invoke them by name in chat. The agent recognizes the skill name and runs the SKILL.md as instructions.
Detect drift between PRD and shipped reality. Surfaces drift in three dimensions: shipped but not in PRD, in PRD but not shipped, invalidated assumptions. Updates with append-only Revision log entry.
Review a single architectural decision against existing ARCHITECTURE.md. Reframes the question, surfaces trade-offs, recommends a path, appends to Trade-off log. Auto-recommended by triage when an item is architecturally loaded.
Run code review, pause for human merge confirmation, execute merge mechanics (rebase + push for worktree, push for direct-on-main), flip status to [x], clean up.
UI vs DESIGN.md. Aesthetic match, forbidden defaults, token discipline, type contrast, spatial rhythm, state coverage, motion sanity. Multi-viewport screenshots if Playwright MCP is available.
After any UI change. Auto-invoked by /craft-ui Phase 8.
Mines transcript deltas. Updates AGENTS.md. Refreshes .claude/hooks/state/continual-learning-index.json. Only writes ## Learned User Preferences and ## Learned Workspace Facts sections. Each section capped at 12 bullets.
Auto-invoked by the continual-learning skill. Or invoke directly via the Agent tool.
Bun TypeScript. Tracks cadence (turns + minutes since last run + transcript advance). On eligible Stop events emits {"decision":"block","reason":"..."} to inject a follow-up that runs the continual-learning skill. Has a loop guard via stop_hook_active. Trial mode via env var.
Bash linter. Fails on raw Tailwind color utilities (bg-red-500, etc.), hex literals in className, forbidden default fonts (Inter, Roboto, Open Sans, Lato, Arial), Material default easing cubic-bezier(0.4, 0, 0.2, 1).
Want to write the PRD? → prd-grillWant to write ARCHITECTURE.md? → architecture-md-builderWant to write DESIGN.md? → design-md-builderWant to refresh the PRD? → prd-reviseWant to decide one architectural question? → architecture-reviewWant to add a forbidden default? → /forbidWant to scaffold a component? → /scaffold-componentWant to build a page or feature UI? → /craft-uiWant to capture an idea fast? → /backlog-intakeWant to triage the Inbox? → /backlog-triageWant to know what to work on next? → /pick-next-taskWant to start work on a non-UI spec? → /kickoff-specWant to ship a finished spec? → /ship-specWant to follow up on deferred ship items? → /ship-followupWant to review a UI change? → invoke design-reviewerWant to review architectural changes? → invoke architecture-reviewerWant to review a schema migration? → invoke migration-reviewerWant to review API endpoints? → invoke api-reviewerWant to refresh AGENTS.md memory? → continual-learning (or wait for cadence)
