What is the Neural Link?

The Neural Link is a bi-directional bridge that connects AI agents directly to the Neo.mjs runtime. It lets agents inspect the Scene Graph, component state, event listeners, computed styles, and DOM rectangles, and mutate the running application in real time.

Why is Neo.mjs called an Application Engine instead of a framework?

Neo.mjs maintains persistent application objects in a worker-backed Scene Graph instead of compiling application state away into ephemeral DOM nodes. That architecture enables multi-window orchestration, runtime permutation, and deep AI introspection.

What is Context Engineering?

Context Engineering shapes the information and tool environment around AI agents. Neo.mjs implements it through Knowledge Base, Memory Core, GitHub Workflow, and Neural Link MCP servers for frontier harnesses, plus a File System MCP server for internal Neo.ai.Agent local loops.

What is the Neo.mjs Agent OS?

The Neo.mjs Agent OS is the repository Brain: source code and services for Memory Core, Knowledge Base, Active Hybrid GraphRAG, DreamService, Golden Path synthesis, A2A coordination, and Neural Link tooling.

Frontmatter

id	17889
title	Migrate the drag sensors to Pointer Events with pointer capture
state	Open
labels	enhancementaiarchitecture
assignees	[]
createdAt	Aug 30, 2026, 8:19 PM
updatedAt	Oct 4, 2026, 1:04 PM
githubUrl	https://github.com/neomjs/neo/issues/17889
author	neo-opus-vega
commentsCount	2
parentIssue	null
subIssues	[]
subIssuesCompleted	0
subIssuesTotal	0
contentTrust
projected
quarantined	0
signals	[]
blockedBy	[]
blocking	[]

Migrate the drag sensors to Pointer Events with pointer capture

Open Backlog/active-chunk-19 enhancementaiarchitecture

neo-opus-vega commented on Aug 30, 2026, 8:19 PM

Context

Two defect classes measured this week both trace to the same foundation: the drag sensors are a mouse/touch-event, hit-testing-based design. (1) A pointer crossing an iframe pane starves the move stream and can swallow pointerup, leaving a session in neo-is-dragging forever — #17883, shielded in PR #17887 by making iframes hit-test-inert during a session. (2) A session whose pointerup is lost terminates never — observed surviving multiple later gestures. Both are symptoms of relying on hit-testing to keep receiving a gesture the sensor already owns.

The platform primitive that removes the class is pointer capture: setPointerCapture(pointerId) routes every subsequent pointer event to the capturing element until release — hit-testing bypassed entirely (iframes included), pointerup/pointercancel delivery guaranteed to the captor, implicit release built in.

The Architectural Reality

src/main/draggable/sensor/Mouse.mjs — claims gestures via testPathInclusion (:135), installs a document-level dragstart → preventDefault for the gesture (:162/:217), stamps body.neo-drag-active (:160/:215); secondary-click/ctrl/meta gating lives here too.
src/main/draggable/sensor/Touch.mjs — a separate sensor for the same job. Pointer Events unify mouse, touch, and pen: the migration plausibly collapses two sensors into one.
src/main/addon/DragDrop.mjs:424 — session cleanup half.
The sensors' OUTPUT contract (drag:start / drag:move / drag:end synthetic events into the worker world) stays untouched — consumers, DragZones, the dock stack, and the #17885 native-drag partition gate never notice.

The Fix

Replace the mouse/touch listeners with pointer listeners; on claiming a gesture, setPointerCapture on a stable capture target; translate to the existing synthetic vocabulary; release capture on end/cancel. Port the existing gating semantics verbatim (dragTargetClasses claim, primary-button/modifier checks, the native-dragstart suppression for claimed gestures, neo-drag-active stamping). Migration details that must not be lost:

touch-action: none (or pan-x/pan-y as appropriate) on drag handles — without it, browser gestures fire pointercancel mid-drag on touch devices.
pointercancel is a first-class terminal now: it must end sessions as cleanly as pointerup (this is the structural fix for the never-terminating session).
The #17887 iframe shield stays in place as defense-in-depth; whether it can retire is a later, evidence-gated decision once capture has soaked (explicitly not part of this ticket).
The #17885 partition contract (native-drag sources exempt from sensor claiming) applies unchanged to the new listeners.

Acceptance Criteria

One pointer-based sensor path drives mouse, touch, and pen; the synthetic drag:* contract is byte-compatible for consumers.
A drag whose pointer crosses an iframe keeps streaming moves and terminates on release inside the iframe — asserted WITHOUT the #17887 shield active in the test (the shield stays shipped; the test isolates the sensor's own robustness).
pointercancel ends a session; no code path can leave neo-is-dragging on a node after the gesture's pointer is gone.
Existing drag/sort/splitter/dock suites green; touch-driven suites (or manual touch verification notes) included.
touch-action styling documented/applied for engine-owned draggable surfaces.

Out of Scope

Retiring the #17887 shield (separate, evidence-gated follow-up).
Native HTML5 drag sources (#17885) — partition unchanged.
The SharedWorkers first-mount race (#17888) — different layer.

#17883 / PR #17887 — the shielded symptom this makes structural
#17885 — the native-drag partition that must keep holding
#17888 — sibling robustness work in the dock stack

Live latest-open sweep: latest 20 open checked 2026-08-30 ~18:20 UTC — no equivalent. A2A claim sweep: none; the idea was routed to @neo-opus-grace via A2A ~18:17 as future-direction — this ticket banks it per operator direction; her #17887 lane is untouched and she has first claim here if she wants it.

Origin Session ID: b51135ac-ec45-4689-ac45-0e99fb073069

Retrieval Hint: "pointer capture drag sensor iframe pointerup stuck neo-drag-active" · query_recent_turns 2026-08-30 (iframe drag forensics)

tobiu unassigned from @neo-opus-grace on Sep 1, 2026, 10:08 AM