feat(phase-3): masterplan refinement (decisions + 10 batches + 30 skill prompts)

- 12 refinement decisions (D-01..D-12) backed by Phase 2 research sections
- 30-skill inventory: 3 core, 10 syntax, 11 impl, 5 errors, 1 agents
- 10-batch execution plan with file-scope-per-worker layout
- Per-skill agent-prompt body (SCOPE bullets + V3_V4_NOTE + SECTIONS) for all 30 skills
- DECISIONS.md adds D-008 (dual v3/v4 columns), D-009 (Phase 4 skip), D-010 (tmux-orchestration)
- ROADMAP advances to 35%

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: FreekHeijting

DECISIONS.md

@@ -64,3 +64,30 @@ Numbered decisions (D-XXX) with rationale. Immutable once recorded — new decis
 - **Decision**: Publish all skill packages under the OpenAEC Foundation GitHub organization.
 - **Rationale**: Centralized, consistent branding. Community ownership. Discoverability.
 - **Consequence**: All repos follow OpenAEC naming conventions and include social preview banners with OpenAEC branding.
+
+---
+
+## D-008: Dual v3/v4 Coverage Per Skill
+
+- **Date**: 2026-05-19
+- **Decision**: Every skill where Tailwind v3.4 and v4.0 diverge MUST present explicit v3 column and v4 column (tables or labeled snippets), never a single shared snippet that hides version differences.
+- **Rationale**: Phase 2 research (L-001 through L-005) showed v4 is more divergent from v3 than the raw masterplan assumed. `@tailwind` directives removed, JS config no longer auto-loaded, `corePlugins`/`safelist`/`separator` removed without replacement, variant stacking order flipped, default border-color + ring-width changed. A skill that picks one version silently misleads users on the other.
+- **Consequence**: 30-skill inventory includes v4-only skills (functional-utilities, 3d-transforms, modern-utilities) and v3-only context in `impl-config-v3`. All shared skills use side-by-side tables.
+
+---
+
+## D-009: Phase 4 Topic-Research Selectively Skipped
+
+- **Date**: 2026-05-19
+- **Decision**: For Tailwind pkg, Phase 4 topic-research is SKIPPED for skills where vooronderzoek-tailwind.md already covers the topic in sufficient depth (most skills), and EXECUTED only for skills needing deeper drill-down (`impl-plugins-custom`, `impl-tailwind-merge`, `impl-migration-v3-v4`, `errors-dynamic-classes`).
+- **Rationale**: BOOTSTRAP-RUNBOOK §6.3 skip-criteria (Docker L-001) authorises this when vooronderzoek is >40 doc-pages-equivalent and well-cited. Tailwind vooronderzoek is 5668 words with 30 verified sources covering every skill's scope directly.
+- **Consequence**: 26 skills go straight to tmux-worker with vooronderzoek-only reference. 4 skills get topic-research first. Total time saved ~1.5h.
+
+---
+
+## D-010: tmux-orchestration For Phase 5
+
+- **Date**: 2026-05-19
+- **Decision**: Phase 5 skill creation uses the `tmux-orchestration` skill with 3 persistent `skill-builder` workers, batch-loop with quality-gate per worker reply.
+- **Rationale**: 30 skills is above the 15-skill threshold from BOOTSTRAP-RUNBOOK §P-012. Persistent workers with file-scope-isolation + QG-loop is the proven pattern (Blender-Bonsai 73 skills, Frappe 61 skills).
+- **Consequence**: Workers spawn in VS Code panels, reply-language Nederlands, QG cadence "every worker reply". File-scope updated per batch via re-instruct.

ROADMAP.md

@@ -1,45 +1,57 @@
-# ROADMAP — Tailwind CSS Skill Package
+# ROADMAP : Tailwind CSS Skill Package
 
 ## Current Status
 
 | Phase | Description | Status | Progress |
 |-------|------------|--------|----------|
-| Phase 1 | Raw Masterplan | ✅ Infrastructure Ready | 50% |
-| Phase 2 | Deep Research (Vooronderzoek) | ⏳ Next | 0% |
-| Phase 3 | Masterplan Refinement | ⏳ Pending | 0% |
-| Phase 4 | Topic-Specific Research | ⏳ Pending | 0% |
-| Phase 5 | Skill Creation | ⏳ Pending | 0% |
-| Phase 6 | Validation | ⏳ Pending | 0% |
-| Phase 7 | Publication | ⏳ Pending | 0% |
+| Phase 1 | Raw Masterplan | Complete | 100% |
+| Phase 2 | Deep Research (Vooronderzoek) | Complete | 100% |
+| Phase 3 | Masterplan Refinement | Complete | 100% |
+| Phase 4 | Topic-Specific Research | In Progress (selectively skipped per D-009) | 0% |
+| Phase 5 | Skill Creation | Pending | 0% |
+| Phase 6 | Validation | Pending | 0% |
+| Phase 7 | Publication | Pending | 0% |
 
-**Overall Progress**: 7% (infrastructure created, methodology in place)
+**Overall Progress**: 35% (research done, refined masterplan + decisions in place, ready for Phase 4+5)
 
 ## Next Steps
 
-1. Complete Phase 1: Create raw masterplan in `docs/masterplan/{{TECH_PREFIX}}-masterplan.md`
-   - Define preliminary skill inventory based on technology knowledge
-   - Estimate skill count per category
-   - Identify key API areas to cover
-2. Begin Phase 2: Deep Research
-   - Research Tailwind CSS comprehensively using SOURCES.md approved URLs
-   - Create `docs/research/vooronderzoek-{{TECH_PREFIX}}.md`
-   - Use WebFetch for latest official documentation
+1. User-checkpoint on refined masterplan (30 skills, 10 batches, D-08..D-12)
+2. Phase 4 : topic-research for 4 selected skills (`impl-plugins-custom`, `impl-tailwind-merge`, `impl-migration-v3-v4`, `errors-dynamic-classes`)
+3. Phase 5 : invoke `tmux-orchestration` skill, spawn 3 `skill-builder` workers, batch-loop with QG per reply across 10 batches
+4. After Phase 5 : STOP per user instruction. Resume at Phase 6 in next session.
 
 ## Skill Summary
 
 | Category | Estimated | Created | Validated |
 |----------|-----------|---------|-----------|
-| syntax/ | TBD | 0 | 0 |
-| impl/ | TBD | 0 | 0 |
-| errors/ | TBD | 0 | 0 |
-| core/ | TBD | 0 | 0 |
-| agents/ | TBD | 0 | 0 |
-| **Total** | **TBD** | **0** | **0** |
+| core/ | 3 | 0 | 0 |
+| syntax/ | 10 | 0 | 0 |
+| impl/ | 11 | 0 | 0 |
+| errors/ | 5 | 0 | 0 |
+| agents/ | 1 | 0 | 0 |
+| **Total** | **30** | **0** | **0** |
 
 ## Changelog
 
-### Phase 1 — Infrastructure (2026-05-19)
+### Phase 1 : Infrastructure (2026-05-19)
 - Repository structure created
 - Core files initialized (CLAUDE.md, ROADMAP.md, REQUIREMENTS.md, DECISIONS.md, SOURCES.md, WAY_OF_WORK.md, LESSONS.md, CHANGELOG.md)
 - Skill category directories created
-- Ready for raw masterplan creation and deep research
+- Raw masterplan populated with 29 candidate topics
+
+### Phase 2 : Deep Research (2026-05-19)
+- 5668-word vooronderzoek covering all 16 scope areas
+- 30 sources WebFetched + verified (Tailwind docs v3+v4, GitHub issues, plugin repos, tailwind-merge)
+- 6 lessons learned (L-001 .. L-006)
+- 9 newly discovered sub-topics flagged for masterplan refinement
+
+### Phase 3 : Masterplan Refinement (2026-05-19)
+- 12 refinement decisions (D-01 .. D-12) recorded in masterplan
+- 4 ADDs (functional-utilities, 3d-transforms, gradients, modern-utilities, dynamic-classes-error)
+- 3 MERGEs (pseudo-elements+state-modifiers, postcss+cli, frameworks-bundle)
+- 1 DROP (standalone engine-model skill)
+- Final inventory : 30 skills across 5 categories
+- Execution plan : 10 batches of 3 skills, dep-chain `core` then `syntax` then `impl` then `errors+agents`
+- Complete agent-prompt body per skill in masterplan
+- DECISIONS.md updated with D-008 (dual v3/v4 coverage), D-009 (Phase 4 skip), D-010 (tmux-orchestration)