Claude-skill-registry feature-build
Orchestrator skill for the complete feature development lifecycle. Coordinates 5 phases - task selection, component design, build loop, analytics setup, and commit/documentation. Use when building any new feature or enhancement that requires multiple steps.
git clone https://github.com/majiayu000/claude-skill-registry
T=$(mktemp -d) && git clone --depth=1 https://github.com/majiayu000/claude-skill-registry "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/data/feature-build" ~/.claude/skills/majiayu000-claude-skill-registry-feature-build && rm -rf "$T"
skills/data/feature-build/SKILL.mdFeature Build Skill
Orchestrates the complete feature development lifecycle with clear phases, entry/exit criteria, and conversation loop patterns. This skill transforms feature requests into shipped, tested, and documented code.
Contract
Inputs:
- Feature request or task from TASKS.md
- Project with existing component structure (components/ui/)
- (Optional) Style guide page for component development
Outputs:
- Working feature verified in browser
- Components exported from barrel file
- PostHog analytics instrumented
- TASKS.md and CHANGELOG.md updated
- Clean git commit(s) with conventional format
Success Criteria:
- All acceptance criteria pass in browser testing
- No console errors or build failures
- Analytics events firing correctly
- Documentation updated
Philosophy
Core Principles
-
Acceptance Criteria First — Define what "done" looks like before writing code. Testable criteria prevent scope creep and enable verification.
-
Components Live in Style Guide — Every UI component starts in the style-guide page. This ensures reusability, documents all states, and provides a living reference.
-
Test What You Build — Use browser tools to verify each step. Don't wait until the end to discover issues.
-
KISS (Keep It Simple, Stupid) — Implement the simplest solution that meets the criteria. Avoid over-engineering.
-
Loop Back, Don't Push Forward — When something doesn't work, return to the appropriate phase rather than patching around issues.
Conversation Loop Model
Features are built through phases that can loop back when issues arise:
┌───────────────────────────────────────────────────────────────────────────────┐ │ │ │ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌─────────────┐ │ │ │ Task │──▶│Component │──▶│ Build │──▶│Analytics │──▶│ Build │ │ │ │Selection │ │ Design │ │ Loop │ │ Setup │ │Verification │ │ │ └──────────┘ └──────────┘ └──────────┘ └──────────┘ └─────────────┘ │ │ ▲ │ │ │ │ │ │ │ ▼ ▼ ▼ ▼ │ │ │ ◀────────────◀─────────────◀──────────────────────────────────│ │ │ (loop back if criteria fail or build errors) │ │ │ │ │ │ ┌──────────────┐ │ │ └────────────────────────────────────────────────────│ Commit & │◀──│ │ (next feature) │ Document │ │ │ └──────────────┘ │ └───────────────────────────────────────────────────────────────────────────────┘
Relationship to Other Skills
- design-system — Reference for UI patterns, component design principles, style-guide template
- feature-gating — Use when building pro/freemium features with PostHog flags
- versioning — Follow semantic versioning for releases
Phase 1: Task Selection
Entry Criteria
- Fresh conversation or user requests a feature
- TASKS.md exists in project
Actions
-
Read TASKS.md — Identify the next item from Current Sprint
Read TASKS.md and identify the next unchecked item in Current Sprint -
Confirm Scope — Clarify with user what's included and excluded
- What exactly should this feature do?
- What should it NOT do?
- Are there edge cases to consider?
-
Break Down if Complex — Split large features into sub-tasks
- Each sub-task should be completable in one conversation
- Sub-tasks should have clear dependencies
-
Define Acceptance Criteria — Write testable criteria
Acceptance Criteria Format
### Feature: [Name] **Acceptance Criteria:** - [ ] User can [specific action] - [ ] System [specific behavior] when [condition] - [ ] Error case: [scenario] displays [message] - [ ] Mobile: [behavior] works on viewport < 768px
Exit Criteria
- Feature scope is clearly defined
- Acceptance criteria are written and agreed
- Sub-tasks identified if feature is complex
Phase 2: Component Design
Entry Criteria
- Feature scope is clear
- Acceptance criteria are defined
Actions
-
Check Style Guide First — Look for existing components that can be reused or extended
Navigate to /style-guide and snapshot to see existing components -
Design in Style Guide — Build new components in the style-guide page
- Components should be isolated and reusable
- Import from
barrel filecomponents/ui/ - Export new components from barrel file
-
Show All States — Document every component state:
- Default/resting
- Hover/focus
- Loading/pending
- Success/complete
- Error/invalid
- Disabled
- Empty (if applicable)
-
Mobile Considerations — Test responsive behavior
- Use PhoneFrame component for mobile mockups if available
- Consider touch targets (min 44px)
Style Guide Section Template
<section id="my-component"> <SectionHeader title="My Component" /> <div className="space-y-6"> <h4 className="text-xs font-medium text-muted-foreground uppercase tracking-wide"> Variants </h4> <div className="flex flex-wrap gap-3"> <MyComponent variant="default" /> <MyComponent variant="secondary" /> </div> <h4 className="text-xs font-medium text-muted-foreground uppercase tracking-wide"> States </h4> <div className="flex flex-wrap gap-3"> <MyComponent loading /> <MyComponent disabled /> <MyComponent error="Error message" /> </div> </div> </section>
Exit Criteria
- Components visible in
/style-guide - All states documented and interactive
- Components exported from
components/ui/index.ts - Responsive behavior verified
Phase 3: Build Loop
Entry Criteria
- Components designed and in style-guide
- Acceptance criteria ready for testing
Actions
3.1 Backend First
Build data layer and API before wiring to UI:
- Database schema — Add migrations if needed
- Service layer — Business logic functions
- API routes — REST endpoints with Zod validation
- Test API — Use browser network panel or curl to verify
3.2 Frontend Integration
Wire components to backend:
- State management — Zustand store or React state
- API hooks — Data fetching with proper loading/error states
- Component wiring — Connect UI to state and actions
3.3 Browser Testing
Test at each step using Cursor browser tools:
1. Navigate to the page being tested 2. Snapshot to get element references 3. Interact with elements (click, type, etc.) 4. Re-snapshot after state changes 5. Verify expected behavior 6. Screenshot if visual verification needed
See
for detailed testing scenarios.references/testing-patterns.md
Testing Checklist by Component Type
Forms:
- Valid submission works
- Validation errors display correctly
- Loading state during submission
- Success feedback shown
- Error handling for API failures
Lists/Tables:
- Items render correctly
- Empty state displays
- Pagination/infinite scroll works
- CRUD operations update UI
- Loading skeleton shows
Modals/Dialogs:
- Opens on trigger
- Closes on backdrop click
- Closes on escape key
- Form inside modal works
- Focus management correct
Navigation:
- Routes load correctly
- Active state shows current page
- Mobile menu works
- Deep links work
Loop Back Triggers
Return to Phase 2 if:
- Component needs additional states not designed
- UI pattern isn't working as expected
- Need to refactor component API
Return to Phase 1 if:
- Scope is unclear or expanding
- Acceptance criteria need revision
- Feature should be split differently
Exit Criteria
- All acceptance criteria pass in browser
- No console errors
- Mobile viewport tested (if applicable)
- Error states handled gracefully
Phase 4: Analytics Setup
Entry Criteria
- Feature is working in browser
- All acceptance criteria pass
Actions
-
Identify Tracking Needs — What user actions matter?
Feature Type Typical Events Form
,{form}_submitted{form}_errorFeature
,{feature}_used{feature}_completedNavigation {page}_viewedPro Feature
,{feature}_attempted{feature}_upgrade_shown -
Implement Events — Add PostHog capture calls
posthog.capture('{feature}_{action}', { feature: 'feature-name', // action-specific properties }) -
Create Dashboard (if appropriate) — Use PostHog MCP tools
Use mcp_posthog-study-bible_dashboard-create for new dashboards Use mcp_posthog-study-bible_insight-create-from-query for insights -
Pro Features — If feature is gated, follow feature-gating skill:
- Create PostHog feature flag
- Implement
hookuseProFeature() - Track
,_attempted
,_upgrade_shown
events_upgrade_clicked
Event Naming Convention
{feature}_{action} Examples: - notes_created - notes_synced - chat_message_sent - upgrade_modal_shown - model_changed
Exit Criteria
- Key user actions instrumented
- Events firing correctly (verify in PostHog)
- Dashboard created (if feature warrants it)
- Pro feature flags configured (if applicable)
Phase 5: Build Verification
Entry Criteria
- Feature is working in browser
- Analytics implemented (or skipped if not applicable)
Actions
-
Run Production Build
npm run build- Must complete with exit code 0
- Fix any TypeScript errors
- Fix any framework-specific build errors (unused exports, client/server boundaries, etc.)
-
Check Linter
npm run lint- Fix any lint errors (warnings acceptable)
-
Verify No Regressions
- If build fails, loop back to Phase 3 to fix issues
- Do NOT proceed to commit until build passes
Loop Back Triggers
Return to Phase 3 if:
- TypeScript errors in new/modified files
- Build errors related to imports or exports
- Client/server component boundary issues
Exit Criteria
-
succeeds with exit code 0npm run build -
passes (no errors)npm run lint - No TypeScript errors
Phase 6: Commit & Document
Entry Criteria
- Feature complete and tested
- Analytics implemented
- Build verification passed
Actions
-
Update TASKS.md
- Move completed items to Completed section
- Add completion date
- Check off sub-tasks
- Add any new backlog items discovered
-
Update CHANGELOG.md
- Add entry under current version or "Unreleased"
- Summarize what was added/changed/fixed
- Reference any breaking changes
-
Stage and Commit
- Use conventional commit format
- Keep commits atomic (one logical change per commit)
Conventional Commit Format
<type>(<scope>): <subject> Types: feat, fix, docs, style, refactor, perf, test, chore, ci, build, revert
Commit Summary Block
Provide at end of feature for easy copy-paste:
### Done **Summary**: [Brief description of what was built] **Commits** (ohmyzsh): gaa && gcmsg "feat(scope): add feature description" gcmsg "fix(scope): fix related issue" gp
TASKS.md Update Pattern
## Completed - [x] Feature name (YYYY-MM-DD) - [x] Sub-task 1 ✓ - [x] Sub-task 2 ✓ - [x] Sub-task 3 ✓
Exit Criteria
- TASKS.md updated with completion status
- CHANGELOG.md updated with feature summary
- Clean commit(s) with conventional format
- Ready for next feature
Quick Reference
Phase Entry/Exit Summary
| Phase | Entry | Exit |
|---|---|---|
| 1. Task Selection | Feature request | Criteria defined |
| 2. Component Design | Criteria clear | Components in style-guide |
| 3. Build Loop | Components ready | All criteria pass in browser |
| 4. Analytics | Feature working | Events instrumented |
| 5. Build Verification | Analytics done | succeeds |
| 6. Commit | Build passes | Docs updated, committed |
Common Commands
# Start dev server npm run dev # Clear Next.js cache if issues rm -rf .next # Rebuild native modules npm rebuild better-sqlite3 # Git workflow (ohmyzsh) gaa && gcmsg "feat(scope): description" gp
Browser Testing Quick Reference
1. mcp_cursor-ide-browser_browser_navigate → load page 2. mcp_cursor-ide-browser_browser_snapshot → get elements 3. mcp_cursor-ide-browser_browser_click/type → interact 4. mcp_cursor-ide-browser_browser_snapshot → verify changes 5. mcp_cursor-ide-browser_browser_take_screenshot → visual check