| name | ui-kit-pipeline |
| description | Full development pipeline for Acronis UI Kit work. Orchestrates the specialized agent team through a phase-gated workflow: Design → Plan → Development → QA → Review → Audit. Use for any non-trivial component work, new components, Web Component fixes, or feature additions. For simple single-file edits, skip to Development directly. Invoke with /ui-kit-pipeline <ComponentName or feature description>.
|
UI Kit Pipeline
Team orchestration skill for the Acronis UI Kit. Drives a 6-phase
gated workflow using the specialized agents in .claude/agents/.
Consult CLAUDE.md for tech stack and conventions.
Read .ai/START_HERE.md before starting.
Invocation
/ui-kit-pipeline <ComponentName|feature> [--scope <type>] [--from <phase>]
Arguments
| Argument | Purpose |
|---|
ComponentName | Pascal-case component (e.g., Select, Table) or feature description |
--scope new | New component — all 6 phases |
--scope wc-fix | Web Component compatibility fix — skip Design, start at Plan |
--scope bug | Bug fix — start at Development |
--scope audit | Audit only — skip to Audit phase |
--from <phase> | Explicit phase override: design|plan|dev|qa|review|audit |
--solo | Force solo subagent mode (no TeamCreate) |
--dry-run | Phase 0 analysis only, print recommended scope and phases |
Phase Classification
Choose the entry point based on work type:
| Work type | Phases |
|---|
| New component | Design → Plan → Dev → QA → Review → Audit |
| WC fix (blockers exist) | Plan → Dev → QA → Review → Audit update |
| Feature addition | Design → Plan → Dev → QA → Review → Audit update |
| Bug fix | Dev → QA → Review |
| Docs/spec only | Dev → Audit update |
| Audit only | Audit |
When in doubt, run Phase 0 analysis: /ui-kit-pipeline Button --dry-run
Phase 0: Component Analysis (always first)
Before spawning any agent, collect context:
- Read .ai/audit/MASTER-AUDIT.md for current component status.
- Read
.ai/audit/components/<name>.md if it exists.
- Check the component's
@status and @wc-blockers JSDoc tags in source.
- Check .ai/CURRENT_TASKS.md for active work on this component.
- Check .ai/PLANS_STATUS.md for existing plans.
Phase 0 Output
================================================================
AvSelect | @status unstable | Audit: yellow | 330 usages
================================================================
Source: src/components/select/select.vue
@status: unstable
@wc-blockers:
- document.addEventListener on mounted (WC blocker)
- string inject key 'select-group' (WC blocker)
Audit: .ai/audit/components/select.md (yellow)
Plan: none
Active task: none
Recommended scope: wc-fix
Phases: Plan → Dev → QA → Review → Audit update
1. Start pipeline (Plan phase)
2. Design phase first (if API changes needed)
3. Dry-run only (stop here)
Solo vs Team Mode
Default to solo subagents (sequential Task calls). Use a team
only when parallel coordination is needed.
| Condition | Mode |
|---|
| Single component, linear phases | Solo subagents |
| Multiple agents needed in parallel (e.g., researcher + architect) | Team |
| Iterative back-and-forth between agents | Team |
| Simple bug fix or docs update | Solo subagent |
See ~/.claude/skills/team-lead/SKILL.md (or project copy) for full
TeamCreate / SendMessage coordination protocol.
Phase 1: Design
When to run: New component, feature with API changes, or WC
architecture requires a new pattern.
Skip when: The component API is unchanged and the approach is
established (e.g., standard WC fix using existing patterns).
Agent: architect
Inputs:
- Phase 0 analysis
- Existing component source (if refactor)
- .ai/audit/MASTER-AUDIT.md
- Reference component:
src/components/button/
Outputs:
.ai/plans/DESIGN-<component>.md — component API, Shadow DOM
approach, inject key design, slot structure, event interface
.ai/plans/adr/ADR-NNN-<decision>.md — for non-obvious choices
Gate 1 Checklist:
On blocker: Send back to architect with specific issue.
Phase 2: Plan
When to run: Always (except pure bug fix or docs-only).
Agent: architect
Inputs:
.ai/plans/DESIGN-<component>.md (if Design ran)
- Phase 0 analysis
@wc-blockers list from component source
Outputs:
.ai/plans/DECOMPOSITION-<component>.md — implementation units:
- List of files to change
- Order of changes (dependencies first)
- Test strategy
- Expected
@wc-blockers state after fix (none)
Gate 2 Checklist:
On blocker: Send back to architect.
Phase 3: Development
When to run: Always.
Agent: developer-react (this is a React repo; developer-vue exists only
for reuse in Vue projects). Both overlay the technology-independent root
developer at ~/.claude/agents/developer/agent.md.
For gated interactive development of a single component, the developer
should invoke the component-dev skill:
/component-dev <ComponentName> --ticket <JIRA-ID>
The component-dev skill drives: spec → implementation → stories →
tests → docs alignment with fix-and-recheck loops.
For broader changes (composables, multiple components, infrastructure),
the developer implements directly following the DECOMPOSITION plan.
Inputs:
.ai/plans/DECOMPOSITION-<component>.md
.ai/plans/DESIGN-<component>.md (if exists)
- Component source at
src/components/<name>/
Outputs:
- Updated
.vue source with @wc-blockers none (if WC fix)
- Updated
@status tag (if status changed)
- Unit tests in
__tests__/<name>.spec.ts
- Stories in
__stories__/<name>.stories.ts
- Spec in
__specs__/<name>.yaml (if spec-driven)
Gate 3 Checklist:
On blocker: Send back to developer with specific issue and file:line.
Phase 4: QA
When to run: Always (except docs-only).
Agent: qa
Inputs:
- Phase 3 output (changed files list)
.ai/plans/DECOMPOSITION-<component>.md
- Component source
Outputs:
.ai/team/<component>/verify/qa.md
Gate 4 Checklist (from qa report):
On blocker: QA FAIL = blocker. Send back to developer.
Phase 5: Review
When to run: Always for non-trivial changes.
Skip when: Pure docs update or single-line bug fix.
Agent: devil-advocate
Inputs:
- All phase artifacts (design, plan, qa report)
- Changed source files
Outputs:
.ai/team/<component>/verify/da-review.md
Gate 5 Checklist:
On blocker: Send specific revision task to owning agent. Re-run
QA after fix if source changed.
Phase 6: Audit
When to run: After any substantive component change. Always after
WC fix or status change.
Agent: researcher (new/updated audit) or qa (quick audit update)
Inputs:
- Final component source
- QA report
- Existing
.ai/audit/components/<name>.md
Outputs:
- Updated
.ai/audit/components/<name>.md
- Updated entry in
.ai/audit/MASTER-AUDIT.md
Audit report structure:
# AvComponentName Audit
**Date:** YYYY-MM-DD
**Status:** 🟢 Green | 🟡 Yellow | 🔴 Red
**@status:** ready|stable|wip|etc.
**@wc-blockers:** none | list
## Summary
One-line verdict.
## WC Compliance
- [ ] No document/window access
- [ ] Symbol inject keys with null fallback
- [ ] No global event bus
- [ ] No Teleport to body
- [ ] @wc-blockers tag accurate
## Test Coverage
- Unit tests: N cases, XX% coverage
- Stories: N exports
## Documentation
- Docs page: exists | missing | stub
- Examples: N demos
## Issues
(if Yellow or Red)
- Issue description → recommended action
## History
- YYYY-MM-DD: description of change
Color thresholds:
| Color | Criteria |
|---|
| 🟢 Green | @wc-blockers none, ≥60% test coverage, docs page exists |
| 🟡 Yellow | 1-2 minor issues: partial WC, low coverage, stub docs |
| 🔴 Red | Zero tests, major WC blockers, no docs, broken |
After audit update, verify .ai/audit/MASTER-AUDIT.md
totals are updated (green/yellow/red counts).
Gate Rules
- Only the orchestrator (you) opens gates. Agents do not advance themselves.
- A
devil-advocate BLOCKED = the gate does not open.
- A QA FAIL caused by the new change = blocker (pre-existing failures are warnings).
- Tech-writer closes each phase with a summary when the work is significant enough to document.
- Send-backs quote the exact blocker and assign to the owning agent.
File Ownership
| Phase | Agent | Writes to |
|---|
| Design | architect | .ai/plans/DESIGN-*.md, .ai/plans/adr/ |
| Plan | architect | .ai/plans/DECOMPOSITION-*.md |
| Development | developer-react | src/components/<name>/, __tests__/, __stories__/, __specs__/ |
| QA | qa | .ai/team/<name>/verify/qa.md |
| Review | devil-advocate | .ai/team/<name>/verify/da-review.md |
| Audit | researcher / qa | .ai/audit/components/<name>.md, .ai/audit/MASTER-AUDIT.md |
Never assign two agents to the same file in parallel.
Team artifact layout (canonical — shared with team-lead)
Per-feature coordination artifacts live under .ai/team/<component>/<phase>/,
where <component> is the work-item slug. Phase folders:
explore · design · build · verify · audit — use only the phases this
pipeline runs (QA → verify/qa.md; the Review gate → verify/da-review.md).
Each phase folder holds:
<agent>.md — that agent's work product (e.g. qa.md)
da-review.md — the devil-advocate's review of that phase. Reviews live
in the phase they gate; there is no separate review/ folder.
summary.md — the tech-writer's phase summary
Durable artifacts live outside the phase folders: DESIGN / DECOMPOSITION / ADRs
in .ai/plans/ (+ .ai/plans/adr/), audit reports in .ai/audit/.
Typical Flows
WC fix for existing component (most common now)
Phase 0: Analysis → identifies blockers
Phase 2: Plan → architect decomposes the fix
Phase 3: Dev → developer fixes blockers, runs component-dev
Phase 4: QA → qa verifies no blockers remain, tests pass
Phase 5: Review → devil-advocate clears
Phase 6: Audit → update .ai/audit/components/<name>.md to green
New component
Phase 0: Analysis → confirm it doesn't exist
Phase 1: Design → architect designs API + WC approach
Phase 2: Plan → architect decomposes implementation
Phase 3: Dev → developer builds with component-dev
Phase 4: QA → qa verifies
Phase 5: Review → devil-advocate clears
Phase 6: Audit → create new .ai/audit/components/<name>.md
Bug fix
Phase 0: Analysis → identify affected files
Phase 3: Dev → developer fixes
Phase 4: QA → qa verifies fix, no regressions
Phase 5: Review → devil-advocate clears (optional for trivial fix)
Audit sweep (batch)
Phase 0: Analysis → pick unaudited/outdated components
Phase 6: Audit → researcher audits each, updates MASTER-AUDIT.md
Progress Display
After each gate, print:
================================================================
AvSelect | Pipeline Progress
================================================================
Phase 1 (Design): SKIPPED (wc-fix scope)
Phase 2 (Plan): PASS → .ai/plans/DECOMPOSITION-select.md
Phase 3 (Development): PASS → @wc-blockers none, tests pass
Phase 4 (QA): PASS → .ai/team/select/verify/qa.md
Phase 5 (Review): IN PROGRESS...
Phase 6 (Audit): --
Next: awaiting devil-advocate review
================================================================
Integration with component-dev
In Phase 3, the developer agent should invoke the component-dev
skill for the gated interactive workflow within their session. The
component-dev skill handles:
- Spec classification (Track A/B/C)
- Story completeness gate
- VR preview
- Behavior test gate
- Docs alignment gate
The pipeline's Gate 3 checklist is the exit condition from
component-dev — all component-dev gates must pass before Gate 3
is cleared.
Cardinal Rule
You (the orchestrator) never implement. Route all code changes
to developer-react, all design artifacts to architect, all reviews to
devil-advocate, all verification to qa, and all audit updates to
researcher. Your only actions are: assign, gate, redirect, escalate.