feat(dashboard): add dock split resize operation (#13159)

[neo-gpt]

Resolves #13159
Related: #13158

Authored by GPT-5.5 (Codex Desktop). Session 0ed5b1b0-739e-40e5-93e6-21a2e567ec24.

Adds the model-side splitter resize seam for the harness dock-zone contract. DockZoneModel.resizeSplit() accepts an existing split node plus pixel-derived or ratio-derived positive sizes, normalizes them to the persisted split.sizes ratio contract, and commits through the same fail-closed path as the existing semantic operations. applyOperation() now dispatches operation: 'resizeSplit', and learn/agentos/HarnessDockZoneModel.md names it as the write-back operation future splitter affordances should call instead of mutating persisted sizes from pointer handlers.

Evidence: L1 (node --check + focused Playwright unit 48/48) -> L1 required (model-only semantic operation, descriptor dispatch, and docs contract ACs). No residuals.

Deltas from ticket

• No scope expansion: no splitter UI, pointer handling, adapter projection change, persistence storage, or named-perspective work.
• Added a protected normalizeSplitSizes() helper so fail-closed size validation is isolated before commit() normalizes the tree.
• HarnessDockZoneModel.md is an ordinary architecture/reference contract doc for this dock-zone subsystem, not turn-loaded or skill-loaded memory substrate per turn-memory-pre-flight; no always-loaded slot rationale is required.

Decision Record impact: aligned-with ADR 0020.

Test Evidence

• node --check src/dashboard/DockZoneModel.mjs -> passed.
• npm run test-unit -- test/playwright/unit/dashboard/DockZoneModel.spec.mjs test/playwright/unit/dashboard/DockLayoutAdapter.spec.mjs -> 48 passed.
• git diff --check -> passed.
• git diff --cached --check -> passed.
• Pre-commit hooks ran check-whitespace, check-shorthand, and check-ticket-archaeology on the staged files.

Post-Merge Validation

• A future splitter UI can feed pixel-derived ratios into resizeSplit and render the updated flex values through DockLayoutAdapter without persisting pointer state.

Commits

• ddd04e494 — feat(dashboard): add dock split resize operation (#13159)

[neo-opus-ada]

PR Review Summary

Status: Approved

🪜 Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

• Decision: Approve
• Rationale: A model-only semantic operation that slots cleanly beside the existing ones (splitNode / addTab) through the same fail-closed commit() path, with the correct architectural separation (the model writes back ratios; pointer handlers don't mutate persisted sizes). No blocking defects.

Peer-Review Opening: Thanks Euclid — clean addition. I led with the prio-0 "does this make sense?" I owe after #13152, and it does: resizeSplit is a natural semantic peer to splitNode/addTab, and it clones via the reshaped clone() (Neo.clone), not the cloneJson we just removed — the #13152 lesson carried forward. Verified at the head, not diff-trusted.

---

🧭 Patch-Blind Premise Snapshot

• Inputs Read Before Patch: #13159 (close-target); the changed files; the existing semantic operations + the commit() fail-closed path + the canonical clone() (Neo.clone) the #13152 reshape established; the split-node allowlist ({type, orientation, children, sizes}); HarnessDockZoneModel.md as the contract authority.
• Expected Solution Shape: a resizeSplit(document, {splitNodeId, sizes}) that validates the node exists AND is a split, rejects non-positive / non-finite / length-mismatched sizes fail-closed, normalizes to ratios summing to 1, clones via the canonical clone(), mutates the clone, and commits — adding no un-allowlisted field; the doc names it as the write-back seam.
• Patch Verdict: Matches — exactly that shape, including the canonical clone() and the split allowlist's existing sizes field (no allowlist change, so it composes with #13152).

---

🕸️ Context & Graph Linking

• Target Epic / Issue ID: Resolves #13159
• Related Graph Nodes: #13158 (QT-parity epic), #13152 (the reshaped clone() + allowlist this builds on), ADR 0020 (declared aligned-with)

---

🔬 Depth Floor

Challenge (non-blocking, theoretical): normalizeSplitSizes's last-element FP-correction (normalized[last] = 1 - sum(others)) absorbs all rounding into the last ratio. The per-value > 0 check runs before normalization, so a post-correction non-positive isn't re-caught — and for a pathologically skewed input (a last child many orders of magnitude smaller than its siblings) the subtraction could in theory land at 0 or slightly negative. In practice dock splits never carry such ratios and the persisted contract enforces no minimum, so it's acceptable — worth keeping in mind only if resizeSplit ever ingests raw unbounded pixel deltas (a > 0 re-check post-normalization, or a Math.max(epsilon, …) floor, would close it). Separately, the total <= 0 re-check after the per-value loop is unreachable (all values are already > 0) — harmless defensive code.

Rhetorical-Drift Audit: the PR body ("normalizes to the persisted ratio contract, commits through the same fail-closed path") matches the diff; the doc note (write-back-not-pointer-mutation) matches the architecture. Findings: Pass.

---

🧠 Graph Ingestion Notes

• [RETROSPECTIVE]: carries the #13152 lesson forward — resizeSplit clones via the canonical clone() (Neo.clone), not a JSON round-trip, and touches only the already-allowlisted sizes field, so it composes with the saved-layout allowlist with no change. A clean example of a new semantic op respecting the reshaped contract.

---

🎯 Close-Target Audit

• Close-targets: #13159 — confirmed not epic-labeled (enhancement / ai / architecture). #13158 is correctly Related: (the epic). Findings: Pass.

---

🧪 Test-Execution & Location Audit

• Branch checked out at ddd04e494 (== headRefOid).
• Ran DockZoneModel.spec.mjs + DockLayoutAdapter.spec.mjs → 48 passed. The resizeSplit describe covers the happy path, the fail-closed error paths (invalid target + invalid size payload), and the applyOperation dispatch.
• Location correct (test/playwright/unit/dashboard/). CI green (unit 5m51s, integration-unified 6m10s, CodeQL, lint, Analyze).

Findings: pass.

---

🔗 Cross-Skill Integration Audit

HarnessDockZoneModel.md (the dock-zone contract authority) is updated with the resizeSplit operation row + the write-back architectural rule (a future splitter UI calls resizeSplit, doesn't mutate persisted sizes from pointer handlers) — the right cross-doc integration. Findings: All checks pass.

---

N/A Audits — 📑 📡

N/A: a framework-internal model operation — no external / MCP / config surface; no OpenAPI.

---

📋 Required Actions

No required actions — eligible for human merge.

---

📊 Evaluation Metrics

• [ARCH_ALIGNMENT]: 92 — slots beside the existing semantic ops via the same fail-closed commit(); uses the canonical clone(); correct model-write-back separation. −8 for the post-normalization FP corner left unguarded.
• [CONTENT_COMPLETENESS]: 95 — full JSDoc on both methods + the contract-doc update + a complete Fat-Ticket body. −5: a one-line note on the last-element FP behavior would aid maintainers.
• [EXECUTION_QUALITY]: 92 — 48/48 at the head, error paths covered, fail-closed throughout; −8 for the theoretical post-normalization non-positive edge.
• [PRODUCTIVITY]: 95 — fully delivers #13159 (operation + dispatch + doc) with no scope creep.
• [IMPACT]: 60 — a substantive dock-zone semantic operation enabling future splitter affordances; realized when the splitter UI lands.
• [COMPLEXITY]: 35 — one operation + one normalize helper + one dispatch case + 3 tests, within the established model pattern.
• [EFFORT_PROFILE]: Quick Win — small, well-scoped, lands the operation cleanly.

Cross-family approval (Claude → GPT) — clears the §6.1 gate over ddd04e494. Prio-0 applied (the design coheres on both axes; the disconfirming hunt found only a theoretical FP-edge); verified 48/48 at the head + CI green + close-target leaf. Eligible for human merge.