Render dock splitters that commit resizeSplit

Context

Epic #13158 identifies interactive splitter resize as the first QT-parity docking polish gap under the Agent Harness line. The current source is ahead of the epic body's oldest wording: DockZoneModel.resizeSplit() and descriptor dispatch already exist, and DockLayoutAdapter.projectSplitNode() already projects persisted split sizes as child flex values.

The remaining near-term gap for this leaf is the adapter/UI seam named in learn/agentos/HarnessDockZoneModel.md Split Projection: resizable splitters should sit between projected children and write back semantic size changes through resizeSplit, not mutate persisted sizes directly from pointer handlers.

Scope correction, 2026-06-14: cross-family review on PR #13209 surfaced that this ticket's original AC wording mixed the adapter/seam proof with the later live Harness-view pointer/drag handler. This ticket is now scoped to the delivered splitter affordance, semantic descriptor helper, and descriptor-to-model commit proof. Live pointer/drag wiring is tracked separately in #13213 under the same parent epic.

Release classification: ON Project 13 Agent Harness board - child leaf under #13158 / #13012, M3 dashboard composition polish.

Live latest-open sweep: checked the latest 20 open issues at 2026-06-14T09:21:19Z; no equivalent leaf found. Exact live searches for DockZoneModel splitter resize, resizeSplit splitter, and dock splitter resize resizeSplit returned no open duplicate beyond parent #13158. A2A in-flight sweep checked the latest 30 messages via a fresh Memory Core client bound as @neo-gpt; no recent [lane-claim] / [lane-intent] overlapped this scope. Follow-up #13213 now owns the live Harness-view pointer/drag interaction.

The Problem

The model contract is ready for persisted split ratios, but the rendered dock layout still needs a stable splitter affordance and semantic helper before the live Harness-view handler can safely exist. Without a scoped adapter seam, a future interaction implementation could cut around the contract by wiring pointer handlers straight to node.sizes, storing transient geometry, or introducing a second docking-specific drag pipeline.

That would violate the #13158 guardrails:

• semantic-operation-only persistence;
• JSON-first model state;
• no DOMRect, pointer coordinates, windowId, preview payloads, functions, or live component references in serialized layout data;
• no parallel drag system.

The leaf should make the first affordance and semantic commit seam real while preserving the already-landed model boundary. The later user-facing drag gesture is a separate Harness-view integration leaf, now #13213.

The Architectural Reality

• learn/agentos/HarnessDockZoneModel.md defines resizeSplit(splitNodeId, sizes) and says future resizable splitters sit between projected children and write through that operation.
• src/dashboard/DockZoneModel.mjs implements resizeSplit(), normalizeSplitSizes(), and applyOperation({operation: 'resizeSplit', ...}) with fail-closed validation.
• test/playwright/unit/dashboard/DockZoneModel.spec.mjs already covers valid resize normalization, invalid resize fail-closed behavior, and descriptor dispatch.
• src/dashboard/DockLayoutAdapter.mjs projects split nodes as ordinary containers with child flex values; this ticket extends that projection with splitter affordance metadata and a pure descriptor helper.

This ticket therefore does not add the model operation or the live pointer/drag handler. It adds the adapter/UI seam and focused commit proof that consumes the existing operation. #13213 owns the later Harness-view interaction path.

The Fix

Add the minimal dashboard-layer splitter affordance seam needed for split resize:

1. Extend the split projection / harness dock rendering path so each adjacent pair of projected split children has a stable splitter affordance with the owning splitNodeId, child boundary index, and orientation available to a later interaction handler.
2. Add a semantic helper that converts resolved child sizes into {operation: 'resizeSplit', splitNodeId, sizes}.
3. Keep transient pointer geometry runtime-only. Pixel deltas may be used by a later handler to compute ratios for the event frame, but only normalized positive sizes enter the dock-zone document.
4. Add focused tests proving the projected affordance metadata exists and that a simulated resolved-size resize commits through resizeSplit without direct sizes mutation.

Contract Ledger Matrix

Decision Record impact

Aligned-with ADR 0020. This is an H1/Harness dashboard composition leaf under the Agent Harness concept; it preserves the ADR guardrails around JSON-first state, benchmark-before-perf-claims, restart-affordance awareness, and dashboard-layer ownership.

Acceptance Criteria

• Split projections expose/render a stable splitter affordance between adjacent split children for both horizontal and vertical split nodes.
• Projected splitter affordances carry the metadata needed by a later handler: splitNodeId, boundary index, and orientation.
• A resolved-size resize can be converted into a resizeSplit semantic descriptor and committed through the existing DockZoneModel.resizeSplit() / applyOperation() path.
• Invalid resize descriptor input fails closed with the original document unchanged.
• Persisted dock-zone / saved-layout data remains JSON-only and contains no pointer coordinates, DOMRect, windowId, preview payloads, functions, or live component references.
• Focused verification covers adapter/affordance metadata, descriptor-to-model commit, source-model non-mutation, and runtime-geometry rejection at the narrowest practical level.

Out of Scope

• Live Harness-view pointer/drag/keyboard interaction wiring for the rendered splitter; tracked in #13213.
• Adding or changing the resizeSplit model operation itself; it already exists.
• Auto-hide / pin sidebars, named perspectives, grouped tab drag, and tab overflow.
• Cross-window choreography, popup terminal drop, or OS-window drag reintegration.
• A core src/layout/Dock primitive or new generic splitter framework.
• Public performance claims before #13032 provides benchmark evidence.

Avoided Traps

• Direct pointer-handler mutation of node.sizes is rejected; the operation seam owns persistence.
• Persisting runtime geometry is rejected; only semantic normalized ratios survive.
• Adding a new docking-specific drag substrate is rejected; use the existing dashboard/drag interaction path or keep the interaction local to the rendered splitter affordance until it commits a semantic operation.
• Re-filing a model-operation ticket is rejected because current source already ships DockZoneModel.resizeSplit().
• Collapsing the live interaction into this seam leaf is rejected after PR #13209 review; the durable follow-up is #13213.

Related

Parent: #13158. Follow-up: #13213. Related: #13012, #13030, #13032. Contract: learn/agentos/HarnessDockZoneModel.md. Substrate: src/dashboard/DockZoneModel.mjs, src/dashboard/DockLayoutAdapter.mjs.

Origin Session ID: 650bf204-359f-41ff-bc35-19c15da38f54

Handoff Retrieval Hints:

• query_summaries("DockZoneModel resizeSplit splitter affordance #13158")
• ask_knowledge_base("DockZoneModel splitter resize sizes semantic operation HarnessDockZoneModel")
• rg -n "resizeSplit|projectSplitNode|Split Projection" learn/agentos/HarnessDockZoneModel.md src/dashboard test/playwright/unit/dashboard