Branch K — Next Prompt Pack (14 prompts)
Branch K — Next Prompt Pack (14 large prompts)
Paste-ready, self-contained macros. Each names its audience (Design / Code / GPT), its prerequisite, and its forbidden list. All UI prompts are render-shell only; all backend prompts are read-first / governed-mutation-only.
1. Claude Design — Unified Canvas visual prototype
MOW_UNIFIED_CANVAS_CLAUDE_DESIGN_VISUAL_PROTOTYPE — Read docs 01/02/04/07 of this pack + Cowork PHU-LUC-A/C. Produce high-fidelity wireframes for the 8 components (doc 07 §8): MowHierarchyCanvas, MowTreeList, MOTCard (4-level progressive disclosure), ProposalOverlay (incl. new reorder + comment/audio), TrafficLightChip (4-state + 9+2), GovernanceCockpit panels, VirtualGrid, WorkflowGraph. Light theme locked; dark variant via token pairs. Resolve T1 label. Forbidden: no code, no backend assumptions, no "Sửa trực tiếp", no color-only status, no mobile-before-desktop.
2. Claude Code — Read-only UI shell on mocked JSON
MOW_CANVAS_CLAUDE_CODE_MOCKED_SHELL — Read docs 03/08. Build the Nuxt render shell driven by mocked CanvasNode[]/CanvasContext exactly matching doc 03. Wire tier tabs, breadcrumb, drill, child-expand, filters, proposal-mode toggle, all empty/loading/error/large-guard states; virtualized 10k-row mock. Forbidden: no business logic, no real backend, no hardcoded data, no raw SQL.
3. Backend — view-contract validation (live PG, read-only)
MOW_CANVAS_BACKEND_VIEW_CONTRACT_VALIDATION_LIVE_PG — Read-only query_pg/pg_schema. Confirm v_mow_design_workflow/v_mow_design_step/v_mow_governance_cockpit (post-EXTEND) project every doc-03 field; benchmark the pre-aggregated v_mow_design_workflow rewrite vs correlated subqueries; verify EXTEND columns + 8 FKs exist (or report pending). Run the reuse-first survey of the 31 logical registries (REUSE/EXTEND/NEW/DEFER). Forbidden: no DDL, no mutation, no apply channel.
4. MOW canvas mock-JSON generator
MOW_CANVAS_MOCK_JSON_GENERATOR — From live read PG (or fixtures), emit realistic CanvasNode[] per tier (T6→T1) + CanvasContext fixtures incl. edge cases: dangling ref (🔴), partial binding (🟡), frozen, 10k-row page, empty tier. Output JSON only for Stage-2 shell. Forbidden: no write, no Nuxt.
5. Proposal-mode DOT integration
MOW_PROPOSAL_MODE_DOT_INTEGRATION — Read docs 05/06. Implement + register dot_mow_design_propose_change (reuse workflow_change_requests); auto-capture node/tier/workflow/step/neighbor context; wire add/edit/delete/reorder forms → draft rows; cockpit diff. Forbidden: no auto-apply, no agent approval, no activation without ≥2 human cross-signs, no optimistic governance update.
6. Governance Cockpit read-only UI
MOW_GOVERNANCE_COCKPIT_READONLY_UI — Read doc 04 Surface 4. Build problem-first cockpit over v_mow_governance_cockpit: pending proposals, approvals-needed (vs ≥2), frozen/failing designs, gate light (fn_iu_gate_verify_closed), severity queue, T6→T1 heatmap. Read-only; approve/reject visible to council only. Forbidden: no agent decision row, no hiding 🔴 items.
7. MOT/JFT task-surface wireframe
MOT_JFT_TASK_SURFACE_WIREFRAME — Read doc 04 Surface 2 + PHU-LUC-D. Refactor the Task Card to 4-level progressive disclosure (L1 name+countdown+9-state badge+action → L2 MOIT form → L3 MOUT table → L4 guide IU+audit). 9-state badges + countdown chips. Forbidden: no runtime mutation (Phase 1 read-only), no direct-save (queue channel only), no code.
8. MOW Design detail-panel wireframe
MOW_DESIGN_DETAIL_PANEL_WIREFRAME — Read doc 04 Surface 3. Wireframe the Design Registry detail: header (owner/version/freeze/health) + step list (bind state, dot_ref, event refs, is_iu_bound) + relation list + Standard Process View (DAG, parallel swimlanes, sub-workflow collapse, zoom/pan/mini-map for 100–500+ steps). Forbidden: no editing outside Proposal Mode, no code.
9. Accessibility review
MOW_UI_ACCESSIBILITY_REVIEW — Audit all surfaces vs WCAG 2.1 AA: color+icon+text triplet, color-blind-safe shapes, contrast pairs (light+dark), keyboard nav (tabs→breadcrumb→filters→roving grid), aria-labels, focus rings. Produce a findings + fix list. Forbidden: no visual redesign beyond a11y, no color changes to the status set.
10. Scale / performance review
MOW_UI_SCALE_PERFORMANCE_REVIEW — Validate 10k-workflow / 20k-governance behavior: keyset pagination (no OFFSET), virtualization, pre-agg counts, indexed filter columns, hard page cap. Load-test the canvas read DOTs. Report latency vs target. Forbidden: no schema change without governance; read-first.
11. WF-001 UI pilot
MOW_WF001_UI_PILOT — Read doc 09 Stage 5 + Phase 1 doc 09. Drive WF-001 (10 steps) through the UI: bind design IU + 10 step IUs/guides + dot_ref/event refs → validate 🟢 → render tree → (council) activate → audit. Prove reversibility (un-bind / rollback to v4). Forbidden: no activation without ≥2 human cross-signs, no irreversible step, no runtime instance.
12. Nuxt render-shell implementation (after design approval)
MOW_NUXT_RENDER_SHELL_IMPLEMENTATION_POST_DESIGN — Prereq: doc-07 design approved + doc-08 contract validated + read DOTs registered. Implement the approved shell against live read DOTs via Directus; capability-driven chips; Điều-28 zero-logic; bind to design_templates rows. Forbidden: no business logic in Nuxt, no direct PG, no hardcoded data, no Qdrant write, no runtime.
13. Reorder + comment/audio proposal design
MOW_PROPOSAL_REORDER_COMMENT_AUDIO_DESIGN — Design the two missing proposal interactions: drag-reorder (capture {order_index}) and comment/audio capture (change_type=comment + optional audio IU attachment). Keep input minimal, context auto-captured. Forbidden: no direct mutation, no new SoT, audio stored as IU ref not inline blob.
14. Tier-table reuse-first ratification packet
MOW_TIER_TABLE_REUSE_FIRST_RATIFICATION — Read-only survey: do tbl_linh_vuc/cong_ty/phong_ban/chuyen_mon/nhiem_vu_type/task_type exist? Map each to REUSE/EXTEND/NEW/DEFER vs existing classification/tenant tables; produce a human ratification packet (no self-approval) for the tier substrate so T6–T4 can leave mock mode. Forbidden: no table creation, no DDL, no self-approval.
Prompt pack verdict
14 delivered (≥12 required). Each prompt is self-contained, names audience + prerequisite + forbidden list, and advances exactly one of: Design prototype, Code shell, backend contract, proposal integration, cockpit, task surface, detail panel, a11y, scale, pilot, Nuxt impl, design gaps, tier substrate.