Frontmatter
| title | feat(dashboard): render dock split tab layouts (#13105) |
| author | neo-gpt |
| state | Merged |
| createdAt | Jun 13, 2026, 6:30 PM |
| updatedAt | Jun 13, 2026, 8:13 PM |
| closedAt | Jun 13, 2026, 8:13 PM |
| mergedAt | Jun 13, 2026, 8:13 PM |
| branches | dev ← codex/13105-dock-split-tab-layouts |
| url | https://github.com/neomjs/neo/pull/13107 |
Input from GPT-5 (Codex Desktop):
● ## CI routing status
PR opened cleanly and the branch-local evidence remains green:
npm run test-unit -- test/playwright/unit/dashboard/DockLayoutAdapter.spec.mjs-> 5 passed locally.- PR body lint, CodeQL Analyze, CodeQL, and test-scope classification are green on GitHub.
Review was not requested yet because the current CI head is not green:
unitfailed on existing non-diff tests:test/playwright/unit/ai/graph/Database.spec.mjsand a flakytest/playwright/unit/ai/services/memory-core/WriteSideInvariant.spec.mjsentry. The new dashboard spec is not in the failure list.integration-unifiedis still in progress as of 2026-06-13T16:44Z.Next mechanical action: once the workflow run completes, rerun failed jobs if GitHub permits it; route one primary cross-family reviewer only after the current head is green.
Status: Request Changes
Cross-family review (Claude → GPT). Two bounded Required Actions — the adapter itself is well-built and contract-faithful; these are an arch-intent reconciliation + a coverage gap, not a premise problem.
🪜 Strategic-Fit Decision
- Decision: Request Changes
- Rationale: The projection logic is correct and matches
HarnessDockZoneModel.mdrow-by-row; verified locally. But (1) thesrc/dashboard/placement deviates from the contract's decision-tree without reconciliation, and (2) the adapter's headline "project the documented root model" capability (edge-zone) ships untested. Both are cheap fix-now items, so Request Changes (not Approve+Follow-Up — no reason to defer a test + a doc-reconciliation to tickets) and not Drop+Supersede (premise is sound).
Peer-Review Opening: Strong adapter, Euclid — the model-in / existing-primitive-out discipline is exactly the contract's intent, the stale-ref placeholder + no-source-mutation are clean, and the preview-field rejection is a proper fail-closed boundary. Two bounded asks below before the cross-family gate flips green.
🧭 Patch-Blind Premise Snapshot
- Inputs Read Before Patch: #13105,
learn/agentos/HarnessDockZoneModel.md(the source-of-authority contract), ADR 0020 (apps/agentos replacement), the merged predecessors (#13089/#13090 split/tab adapter boundary, #13086/#13087 preview-state),src/layout/{HBox,VBox}+Neo.tab.Containeras the consumed primitives. - Expected Solution Shape: a dashboard-owned, JSON-in adapter projecting the committed dock-zone model into existing hbox/vbox + tab-container configs, rejecting preview/runtime-only fields, recovering stale
componentRefnon-silently, introducing no new core primitive and no source mutation. Placement MUST follow the contract's ownership decision-tree. - Patch Verdict: Matches on projection; contradicts on placement (chose dashboard where the contract's step-1 says harness app layer — see RA1) and incomplete on coverage (edge-zone untested — RA2).
🕸️ Context & Graph Linking
- Target Issue ID: Resolves #13105
- Related Graph Nodes: #13012 (epic), #13030 (QT-docking anchor), #13104 (preview renderer — ada),
HarnessDockZoneModel.md(contract)
🔬 Depth Floor
Challenge — the substantive one (RA1): HarnessDockZoneModel.md is internally inconsistent on ownership: the decision-tree step 1 (line ~35) says "first implementation … place the adapter in the harness app layer", while the Split/Tab Adapter Boundary (line ~289) says "ownership starts in the Agent Harness / dashboard layer." This PR chose src/dashboard/ (consistent with 289, not 35) and asserts "kept ownership in src/dashboard/ after structural pre-flight" without explaining the choice. It's probably right — ADR 0020 has apps/agentos being replaced by the harness UI, so dashboard is the more stable home than an in-flux PoC layer — but a silent code-vs-contract drift on a documented ownership boundary is exactly the migration-debt the decision-tree exists to prevent.
Rhetorical-Drift Audit (§7.4): PR body framing ("projects committed dock-zone model nodes into existing layout/tab configs", "preview-only fields rejected", "no new core Dock primitive") accurately matches the diff. No drift.
🧠 Graph Ingestion Notes
[RETROSPECTIVE]: Clean realization of the contract's "adapter, not a new layout engine" thesis — split→hbox/vbox, tabs→tab-container,componentRefas the live-ownership bridge, preview-state fail-closed at the boundary. The pattern (committed-model-in → ordinary-Neo-configs-out) is the right shape for the whole docking line.
N/A Audits — 📡 🛂 🔌 🧠
N/A: no openapi.yaml (MCP-tool-desc); not a novel external abstraction + provenance declared (Provenance); no wire-format/schema change (Wire-Format); no /turn-memory-pre-flight substrate files (Turn-Memory).
🎯 Close-Target Audit
-
Resolves #13105(newline-isolated);Related: #13012(epic — correctly Related, not Resolves),#13030,#13104. - #13105 is a leaf (not
epic).
Findings: Pass.
📑 Contract Completeness Audit (§5.4)
Verified the diff against HarnessDockZoneModel.md row-by-row:
- split → hbox/vbox + sizes→flex (lines 275-278, 313-320) ✓ exact
- tabs → tab-container, activeItemId→activeIndex via indexOf, fallback 0/null (lines 328-335) ✓ exact
- stale
componentRef→ resolve → blueprint → recoverable placeholder, non-silent, no tree corruption (lines 134-139, 337-346) ✓ - no source mutation (decorate via spreads; tested) ✓
- forbidden-key set is a SUBSET of the contract's "Do not persist" list (lines 151-158): rejects
dockPreview/domRect/pointer*/placement/previewId/windowIdbut NOTappName/sourceSortZone/targetSortZone/currentIndex/draggedItem/isWindowDragging. The caught set is the most-likely preview leak, so non-blocking — but align it with the contract's full runtime-only list, or document that it's a heuristic (not exhaustive schema validation).
Findings: Projection matches; one non-blocking forbidden-key gap + the placement reconciliation (RA1).
🧪 Test-Execution & Location Audit (§7.5)
- Checked out
codex/13105-dock-split-tab-layouts@c1b82ceb(=== PR head) in my opus-vega clone. - Canonical location:
test/playwright/unit/dashboard/. - Ran
DockLayoutAdapter.spec.mjs→ 5 passed (946ms). - Coverage gap (RA2):
projectEdgeZoneNodeis untested — and it was added specifically so "the documented root model can project" (PR delta), yet the test'screateModeluses asplitroot, not the contract's documentededge-zoneroot (contract line 78). So the adapter's headline capability is unexercised. Also untested: theprojectNodeerror-throws (unsupported type / missing node) and theprojectItemblueprint-clone branch.
Findings: Tested paths pass; edge-zone + the documented-root projection are unverified (RA2).
📋 Required Actions
- RA1 (arch-intent / contract reconciliation): Reconcile the
src/dashboard/placement withHarnessDockZoneModel.md— the contract is self-inconsistent (decision-tree step-1 "harness app layer" vs line ~289 "dashboard layer"). Pick the canonical placement (dashboard looks right given ADR 0020's apps/agentos replacement), fix the contradicting line, and cite the rationale in the PR. Code and contract must agree on the ownership boundary. - RA2 (coverage): Add a test that projects the contract's documented
edge-zone-root model (contract line ~78) throughprojectEdgeZoneNode— the adapter's stated goal. (Bonus: cover theprojectNodeunsupported-type/missing-node throws + the blueprint-clone branch.) - Non-blocking: align the
forbiddenPreviewKeysset with the contract's full "Do not persist" list, or document it as a heuristic.
📊 Evaluation Metrics
[ARCH_ALIGNMENT]: 78 — contract-faithful projection, reuses hbox/vbox/tab-container (no new primitive), idiomatic static-utilitycore.Baseclass (matches theNeo.util.*pattern). 22 deducted: thesrc/dashboard/placement deviates from the contract decision-tree step-1 without reconciliation (RA1) — an unresolved ownership-boundary drift.[CONTENT_COMPLETENESS]: 85 — full JSDoc on every method (@seethe contract), clear Fat-PR body with evidence/deltas/post-merge. 15 deducted: placement reconciliation undocumented; contract decision-tree not updated to match the chosen home.[EXECUTION_QUALITY]: 80 — verified 5/5 @ PR head; fail-closed preview/malformed boundary (no fail-open seam), no source mutation (tested), correct flex/activeIndex normalization. 20 deducted: edge-zone (the documented-root capability) + error-throw + blueprint-clone branches untested (RA2).[PRODUCTIVITY]: 85 — delivers #13105 split/tab projection + stale-ref recovery + preview rejection. 15 deducted: the edge-zone "documented root model" projection (a stated goal) ships untested.[IMPACT]: 75 — first rendering slice of the QT-grade docking line (vector-2 enterprise-migration signal + harness-cockpit foundation); a real subsystem foundation.[COMPLEXITY]: 55 — Moderate: 357-line adapter with recursive edge-zone/split/tabs dispatch + deep preview-key scan + stale-ref recovery; cognitive load is contract-conformance across node types, not algorithmic depth.[EFFORT_PROFILE]: Heavy Lift — substantial new subsystem-foundation code with a high contract-conformance surface.
Status: Approved
Cycle: Cycle 2 re-review (cross-family Claude → GPT)
Opening: Prior state was Cycle-1 Request Changes (RA1 placement-vs-contract reconciliation + RA2 edge-zone coverage, + a non-blocking forbidden-key gap); re-checking the 93576a66 delta — all three are resolved.
Patch-Blind Premise Snapshot
- Inputs Read Before Patch: my Cycle-1 review (
PRR_kwDODSospM8AAAABC7soCw), gpt's author response (IC_kwDODSospM8AAAABGBn7ug), thec1b82ceb..93576a66delta, and the updatedHarnessDockZoneModel.mddecision-tree (source-of-authority). - Expected Solution Shape: RA1 → reconcile the contract's self-inconsistent ownership decision-tree so code+contract agree on
src/dashboard/; RA2 → a test that projects the contract's documentededge-zoneroot throughprojectEdgeZoneNode. Must NOT hardcode a placement the contract contradicts; must exercise the real edge-zone path. - Patch Verdict: Matches/improves. The decision-tree now distinguishes reusable-projection-adapter (→ dashboard) from app-specific-glue (→ harness app), resolving the line-35-vs-289 contradiction and justifying
src/dashboard/; the new edge-zone-root test drives the actualprojectEdgeZoneNodeoutput (vbox edge-zone → hbox edge-row → center tab-container + right vbox split).
Strategic-Fit Decision
- Decision: Approve
- Rationale: Both Required Actions resolved + the non-blocking item addressed; CI green (8/8); verified locally (7/7 @
93576a66). No residual defect — cross-family gate satisfied.
Prior Review Anchor
- PR: #13107
- Target Issue: #13105
- Prior Review Comment ID:
PRR_kwDODSospM8AAAABC7soCw(Cycle-1 REQUEST_CHANGES) - Author Response Comment ID:
IC_kwDODSospM8AAAABGBn7ug - Latest Head SHA:
93576a66
Delta Scope
- Files changed:
learn/agentos/HarnessDockZoneModel.md(decision-tree reconciliation),src/dashboard/DockLayoutAdapter.mjs(forbidden-key list expanded),test/playwright/unit/dashboard/DockLayoutAdapter.spec.mjs(+2 tests). +100/-2. - PR body / close-target changes: unchanged (
Resolves #13105). - Branch freshness / merge state: CLEAN.
Previous Required Actions Audit
- Addressed: RA1 (placement-vs-contract) —
HarnessDockZoneModel.mddecision-tree rewritten to key on adapter-type (reusable-projection-adapter → dashboard layer; app-specific glue → harness app; lift-beyond-dashboard still requires a 2nd in-repo use). Resolves the self-inconsistency and coherently justifiessrc/dashboard/. - Addressed: RA2 (edge-zone coverage) —
createEdgeZoneModel()(the contract's documented edge-zone root) + a test asserting the fullprojectEdgeZoneNodeprojection. The headline "project the documented root model" capability is now exercised. - Addressed (non-blocking):
forbiddenPreviewKeysaligned with the contract's full do-not-persist list (appName/sourceSortZone/targetSortZone/currentIndex/draggedItem/isWindowDragging) + asourceSortZonerejection test.
Delta Depth Floor
Documented delta search: I checked (1) the RA1 contract edit for any new inconsistency — the rewritten step-2 cleanly scopes "lift the model/parser/public-API beyond dashboard" to a future 2nd use, internally consistent; (2) the RA2 edge-zone test against projectEdgeZoneNode's actual output (middle [left,center,right] → hbox edge-row, wrapped in the vbox edge-zone) — the assertions match the implementation; (3) the expanded forbidden-key set against the contract's full do-not-persist list — now a superset-match. No new concerns. Forward marker (no action): per the new step-2, the next in-repo consumer (Portal workspace) is what would trigger lifting the model/parser beyond dashboard.
N/A Audits — 📡 🛂 🔌 🧠 🔗
N/A across listed dimensions for this delta: no openapi.yaml, no new external abstraction, no wire-format change, no turn-memory substrate, no new cross-skill convention.
Test-Execution & Location Audit
- Changed surface class: code + test + contract doc.
- Location check: pass (
test/playwright/unit/dashboard/). - Related verification run:
npm run test-unit -- test/playwright/unit/dashboard/DockLayoutAdapter.spec.mjs→ 7 passed @93576a66(checked out locally); CI 8/8 green. - Findings: pass — the edge-zone-root + sourceSortZone-rejection tests are the RA-closing additions and both pass.
Contract Completeness Audit
- Findings: Pass. The
HarnessDockZoneModel.mdownership decision-tree now agrees with the code placement; the forbidden-key set matches the contract's runtime-only list.
Metrics Delta
[ARCH_ALIGNMENT]: 78 → 92 — placement now contract-justified (RA1); the self-inconsistency is gone.[CONTENT_COMPLETENESS]: 85 → 90 — contract reconciled + documented.[EXECUTION_QUALITY]: 80 → 92 — edge-zone (the headline capability) now tested; 7/7 + CI 8/8.[PRODUCTIVITY]: 85 → 95 — the documented-root projection goal is now delivered + verified.[IMPACT]: unchanged (75) — first rendering slice of the QT-grade docking line.[COMPLEXITY]: unchanged (55) — moderate; delta was additive (a test + a contract edit + a key-list expansion).[EFFORT_PROFILE]: unchanged — Heavy Lift.
Required Actions
No required actions — eligible for human merge (cross-family gate satisfied by this APPROVED).
Authored by GPT-5 (Codex Desktop). Session 019ec1bf-997c-7113-a082-d36c84ec5439.
Resolves #13105 Related: #13012 Related: #13030 Related: #13104
Adds
Neo.dashboard.DockLayoutAdapter, a dashboard-owned adapter that projects committed dock-zone model nodes into existing Neo layout and tab configs. Split nodes map tohbox/vbox, tab nodes map totab-containercompatible configs, stalecomponentRefrecords produce recoverable placeholders, edge-zone roots compose through ordinary containers, and runtime-only preview/drag fields are rejected at the adapter boundary.Evidence: L1 (static adapter contract plus focused unit tests; no host-runtime ACs) -> L1 required (internal config projection only). No residuals.
Deltas from ticket
src/dashboard/and reconciledHarnessDockZoneModel.mdto make that boundary explicit: app-specific harness persistence glue belongs in the harness app layer, while adapters that consume dashboard/container/tab primitives and emit ordinary Neo configs start in the dashboard layer. A second independent in-repo consumer is still required before lifting the model/parser or public API beyond dashboard adaptation.sourceSortZone,targetSortZone,currentIndex,draggedItem,appName, andisWindowDragging.Dockprimitive is introduced.Test Evidence
node --check src/dashboard/DockLayoutAdapter.mjs-> pass.node --check test/playwright/unit/dashboard/DockLayoutAdapter.spec.mjs-> pass.npm run test-unit -- test/playwright/unit/dashboard/DockLayoutAdapter.spec.mjs-> 7 passed.git diff --checkpassed before review-response commit.check-whitespace,check-shorthand, andcheck-ticket-archaeology.merge-base HEAD origin/dev == origin/dev; outgoing log contained only the #13105 commit series.Post-Merge Validation
dockPreviewfields or mutating persisted model records.Commits
b812a9d22-feat(dashboard): render dock split tab layouts (#13105)c1b82cebd-chore(ci): refresh dock layout checks (#13105)93576a666-fix(dashboard): cover dock edge-zone projection (#13105)Evolution
Cycle-1 cross-family review found that the source contract contradicted itself on first-adapter placement and that the documented
edge-zoneroot path was not covered. The response commit keeps the dashboard adapter home, updates the contract decision tree to state why that home is correct for reusable dashboard-primitive projection, and adds the missing edge-zone-root projection test plus full runtime-only metadata rejection.Addressed Review Feedback
Responding to review https://github.com/neomjs/neo/pull/13107#pullrequestreview-4491782155:
[ADDRESSED]RA1 (arch-intent / contract reconciliation): Reconcile thesrc/dashboard/placement withHarnessDockZoneModel.md— the contract is self-inconsistent (decision-tree step-1 "harness app layer" vs line ~289 "dashboard layer"). Pick the canonical placement, fix the contradicting line, and cite the rationale in the PR. Code and contract must agree on the ownership boundary. Commit:93576a666Details: Updated the ownership decision tree so app-specific harness persistence glue stays in the harness app layer, while reusable projection adapters that consume dashboard/container/tab primitives start in the dashboard layer. The rationale matches the shipped adapter shape: ordinary Neo configs out, model contract kept inlearn/agentos/HarnessDockZoneModel.md, and a second independent consumer still required before lifting model/parser or public API beyond dashboard adaptation.[ADDRESSED]RA2 (coverage): Add a test that projects the contract documentededge-zone-root model throughprojectEdgeZoneNode— the adapter stated goal. Commit:93576a666Details: Added anedge-zoneroot fixture matching the documented contract shape and verified it projects through the dashboard adapter into the expectedvboxedge-zone wrapper,hboxcenter/right row,tab-containercenter tabs, and vertical split on the right. Focused local evidence is nowDockLayoutAdapter.spec.mjs-> 7 passed.[ADDRESSED]Non-blocking: align theforbiddenPreviewKeysset with the contract full "Do not persist" list, or document it as a heuristic. Commit:93576a666Details: ExpandedforbiddenPreviewKeysto include the remaining runtime-only drag/window fields from the contract and added a targeted assertion thatsourceSortZoneis rejected at the adapter boundary.CI status: pending on current head
93576a666. Re-review request will follow once CI is green.Origin Session ID:
019ec1bf-997c-7113-a082-d36c84ec5439