UI Fullflow Orchestrator

Goal: upgrade 17 existing skills from "collaborative" to "serially executable".

SSOT

• •../contracts/fullflow-handoff.md
• •../contracts/quality-gates.md
• •../contracts/reasoning-visibility.md
• •../README.md
• •../ui-codegen-master/SKILL.md
• •../style-scope-guard/SKILL.md
• •../svg-canvas-icon-engine/SKILL.md
• •../../AI_UI_Skill_Design_Document.md
• •Frontend development specification document (Chinese source file in repo)

One-Click Pipeline Script

bash skills/ui-fullflow-orchestrator/scripts/run_fullflow_pipeline.sh \
  --brief "SaaS dashboard with data readability and strong primary CTA" \
  --style-target "Dashboard header and core card area" \
  --scope-file "src/pages/dashboard.css" \
  --icon-mode auto \
  --icon-style outline \
  --framework react \
  --project-type saas-modern \
  --priority performance \
  --priority accessibility \
  --density comfortable

Default output directory: .fullflow-output/<timestamp>/

Artifacts:

• •flow.input.json
• •requirements.summary.json
• •requirements.prd.md
• •requirements.questions.md
• •style-profile.yaml
• •style.scope.lock.json
• •style.scope.checklist.md
• •icon.manifest.json (on demand)
• •icon.spec.md (on demand)
• •icon.sprite.svg (on demand)
• •selector.recommend.json
• •selector.evaluate.json
• •aesthetic.score.json
• •tokens.json
• •tokens.css
• •stage.status.json
• •quality.gates.md
• •self-eval.scorecard.json
• •optimization.plan.md
• •decision.trace.md (user-visible decision trace summary)
• •fullflow.report.md

Orchestration Flow

Follow the five-phase lifecycle from meta standards:

• •Phase 1 requirement analysis and design exploration
• •Phase 2 architecture planning and component design
• •Phase 3 code generation and implementation
• •Phase 4 self-review and refactor
• •Phase 5 acceptance and delivery

1. •app-workspace-guide: read current workspace baseline
2. •requirements-elicitation-engine: generate requirement draft, question list, style profile
3. •style-scope-guard: lock style-only change boundary and output scope lock
4. •ui-selector-pro: generate recommendation shortlist
5. •ui-selector-pro: run quantified evaluation for shortlist
6. •ui-aesthetic-coach: score brief aesthetics and suggest direction
7. •ui-aesthetic-coach: generate design tokens (JSON/CSS)
8. •svg-canvas-icon-engine: generate icon-system artifacts on demand
9. •ui-generation-workflow-runner: enter implementation
10. •ui-component-extractor + frontend-standards-enforcer: structure and standards optimization
11. •ui-acceptance-auditor + ui-self-reviewer: quality closure
12. •ui-codegen-master: final delivery convergence

Rules

• •Pipeline outputs must follow fullflow-handoff.md.
• •Stage status and quality gates must follow quality-gates.md.
• •Reasoning visibility must follow reasoning-visibility.md: output auditable decision summaries only, not private reasoning details.
• •If required inputs are missing (brief/framework/project-type), fail immediately with clear prompt.
• •If style_target is missing or style.scope.lock.json is not successfully locked, fail immediately.
• •When icon-mode=auto|on and icon requirements are detected, icon.manifest.json must be generated.
• •If requirements.summary.json has completeness_score < 70, mark requirement-completeness risk in report.
• •Must output quantitative scorecard (self-eval.scorecard.json) and optimization plan (optimization.plan.md) as iteration entry.
• •If script output conflicts with document heuristics, keep script output first and explain manual correction rationale in report.

Output Contract

1. •Pipeline execution result (success/failure)
2. •Artifact directory and report path
3. •Recommended library + recommended direction
4. •Stage status summary (five phases)
5. •Next skill chain to execute