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	15212
title	Repair stale FM cockpit drill/a11y E2E witnesses after native-button convergence
state	Closed
labels	bugaitestingregression
assignees	neo-gpt
createdAt	Jul 16, 2026, 9:22 AM
updatedAt	6:51 AM
githubUrl	https://github.com/neomjs/neo/issues/15212
author	neo-fable
commentsCount	4
parentIssue	null
subIssues	[]
subIssuesCompleted	0
subIssuesTotal	0
contentTrust
projected
quarantined	0
signals	[]
blockedBy	[]
blocking	[ ] 15251 Fleet cockpit consumes the dock tear-out seam, [ ] 14613 FM cockpit drill e2e: card → detail → pop-out → reattach round-trip
closedAt	6:51 AM

Repair stale FM cockpit drill/a11y E2E witnesses after native-button convergence

Closed Backlog/active-chunk-6 bugaitestingregression

neo-fable commented on Jul 16, 2026, 9:22 AM

Context

The Fleet cockpit has two red Whitebox E2E witnesses on current dev, but fresh falsification shows that the production card→detail path is not regressed. The ticket's original diagnosis predated the deliberate #14619 / PR #15094 accessibility convergence and therefore prescribed restoring a retired interaction.

Clean-current-dev evidence at 5738546bf4 (2026-07-16):

NEO_E2E_PORT=8212 NEO_TEST_SKIP_CI=true npx playwright test \
  test/playwright/e2e/agentos/FleetCockpitDrillNL.spec.mjs \
  test/playwright/e2e/agentos/FleetGridKeyboardA11y.spec.mjs \
  -c test/playwright/playwright.config.e2e.mjs --workers=1
<h1 class="neo-h1" data-record-id="3">2 failed</h1>

FleetCockpitDrillNL.spec.mjs still clicks .fm-card-avatar. PR #15094 intentionally made each card a non-interactive role=listitem and moved drill activation onto the dedicated native .fm-card-drill Button. The avatar doing nothing is now correct.
FleetGridKeyboardA11y.spec.mjs successfully executes native Enter and Space and renders Bravo, then Charlie. It fails later because it expects a persistent .fm-card-control-status after a successful lifecycle settle. The current lifecycle contract clears pendingAction/controlReason on success, so absence of a persistent status is correct. The test needs deterministic request evidence instead of inferring intent delivery from a state that is intentionally transient.

The prior whole-card restoration experiment is quarantined and must not be revived: it would merge the non-interactive listitem and its three native child Buttons back into competing activation authorities, undoing #14619's accepted semantics.

The Problem

The committed E2E witnesses encode two stale assumptions:

The old avatar/body click target still owns drill activation.
A successful lifecycle intent leaves a persistent control-status marker.

Because custom E2E configurations are not hosted-CI gates, these stale witnesses remained red while the production contracts evolved correctly.

The Architectural Reality

AgentCard is a non-interactive role=listitem. The resident name is a dedicated native drill Button (.fm-card-drill) with native Enter/Space semantics; lifecycle toggle/restart are separate sibling Buttons.
AgentCardController.onCardSelect is intentionally scoped to the drill Button. Restoring a card-root click handler would create competing drill/control authority and regress the accessibility topology ratified in #14619 / PR #15094.
The keyboard witness may execute a lifecycle Button, but it must do so against a deterministic test-owned Fleet bridge. It must prove one minimal request plus no drill leakage without touching a live/shared fleet or relying on transient UI state.
Whitebox evidence remains Playwright interaction → Neural Link engine truth. No direct controller shortcut may replace the real DOM activation path.

The Fix

Update FleetCockpitDrillNL.spec.mjs to click the exact selected card's native .fm-card-drill Button. Keep the exact resident-id possession assertion and update the stale whole-card narrative.
Make the lifecycle-isolation leg in FleetGridKeyboardA11y.spec.mjs deterministic: use a test-owned recording/rejecting Fleet bridge, execute the native lifecycle Button, assert exactly one minimal lifecycle request, assert the named rejection/status without a roster refresh, and prove the detail pane remains on Charlie.
Re-run both scoped Whitebox E2Es together on a clean current-dev baseline.

Contract Ledger

Target Surface	Source of Authority	Required Behavior	Forbidden Fallback	Evidence
Card→detail drill witness	#14619 / PR #15094 native Button convergence	Click the exact resident's `.fm-card-drill` Button and possess that resident in `AgentDetail`	Avatar/body click or direct controller invocation	`FleetCockpitDrillNL.spec.mjs`
Keyboard drill witness	#14619 / PR #15094	Native Enter + Space drill activation remains executable	Restoring an interactive card root	`FleetGridKeyboardA11y.spec.mjs`
Lifecycle isolation witness	B4÷C2 intent seam + #14619 sibling controls	Native control activation emits one minimal request and does not drill	Shared/live fleet mutation; persistent-success-status assumption	Recording loopback + mounted DOM/Neural Link assertions
Production code	Current `AgentCard` / `AgentCardController`	Remains unchanged unless a new falsifier proves a product defect	Whole-card listener restoration	Exact diff + both E2Es green

Acceptance Criteria

FleetCockpitDrillNL.spec.mjs passes by activating the dedicated native drill Button and still proves exact resident possession.
FleetGridKeyboardA11y.spec.mjs passes with native Enter/Space, deterministic lifecycle-request evidence, no drill leakage, focus restoration, zero scroll, and zero uncaught page errors.
The lifecycle leg cannot mutate a live/shared fleet and does not depend on a persistent success status.
No production AgentCard, AgentCardController, or Fleet interaction semantics are changed.
The PR body names both stale assumptions and cites #14619 / PR #15094 as the successor authority.

Out of Scope

Restoring whole-card/avatar activation · changing lifecycle product semantics · changing hosted-CI E2E coverage · the broader drill-tour leaf #14613.

Parent surface #14560 · original detail witness #14608 / PR #15063 · successor accessibility authority #14619 / PR #15094 · broader drill journey #14613.

Successor-risk verdict: the ticket goal remains positive-ROI, but the original production-regression premise and unchanged-spec prescription were superseded by #14619. Amended to valid-as-written at this body revision.

Origin Session IDs: 2c0a23e9-f468-4de6-9e29-ddec96103fb4, d99146da-0478-4f23-bc16-dff04f5d650c, 1200dbd1-5159-45be-84c3-b7e323ad4979

Retrieval Hint: "FM cockpit stale drill witness native Button lifecycle status loopback"