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	15460
title	Card lifecycle witness: the Stop verb completes the power-cycle round-trip
state	Closed
labels	enhancementaitesting
assignees	neo-opus-vega
createdAt	Jul 18, 2026, 1:39 PM
updatedAt	Jul 18, 2026, 5:11 PM
githubUrl	https://github.com/neomjs/neo/issues/15460
author	neo-opus-vega
commentsCount	0
parentIssue	null
subIssues	[]
subIssuesCompleted	0
subIssuesTotal	0
contentTrust
projected
quarantined	0
signals	[]
blockedBy	[]
blocking	[]
closedAt	Jul 18, 2026, 5:11 PM

Card lifecycle witness: the Stop verb completes the power-cycle round-trip

Closed Backlog/active-chunk-7 enhancementaitesting

neo-opus-vega commented on Jul 18, 2026, 1:39 PM

Context

Delivered-slice follow-up from @neo-kimi-phoebe's PR #15445 review (her §7.1 Depth-Floor challenge, explicitly non-blocking and named "the cheap sibling for the next pass"): FleetCardLifecycleNL.spec.mjs witnesses Start and rejected-Start, but the Stop verb — which rides the SAME contextual power toggle and the SAME stateful registry double — has no e2e round-trip witness. The card contract renders exactly one power verb per state (start-when-off, stop-when-running); today only half of that contextual cycle is executed end-to-end.

The Problem

The #15242 retirement re-homed the lifecycle surface onto the cards, and #15439 (PR #15445, merged) restored the e2e tier for the Start leg with the discriminating property: Body state advances only from RE-POLLED REGISTRY truth, never from the click. Stop shares every seam (B4 control-toggle → lifecycleIntent → C2 fleetLifecycleIntentAdapter → the authenticated bridge → settle re-poll), but nothing executes it on the mounted composition — a regression on the Stop wire shape or on the running→stopped re-poll render would ship invisibly.

The Architectural Reality

test/playwright/e2e/agentos/FleetCardLifecycleNL.spec.mjs — the merged witness: stateful loopback bridge (running Set; mutate-on-verb, serve-current-on-read), expectMinimalLifecyclePayload (currently hard-coded to startAgent), bootWiredCockpit, getRosterRecord.
apps/agentos/view/fleet/AgentCard.mjs — ONE contextual toggle: fa-play at state off, fa-stop otherwise; onToggleLifecycle derives the verb from the rendered state.
apps/agentos/view/fleet/fleetLifecycleIntentAdapter.mjs — stop → stopAgent is already a registered mapping (LIFECYCLE_ACTION_METHODS), and stopAgent is in FLEET_WIRE_METHODS.

The Fix

Extend the happy-path test into the FULL power cycle (no new boot, no new spec):

Bridge dispatch gains a stopAgent handler mirroring startAgent: running.delete(request.params) → {ok: true, result: {id: request.params, state: 'stopped'}}.
expectMinimalLifecyclePayload(request, method = 'startAgent') — parameterize the expected wire method.
After the existing Start assertions: click the now-Stop toggle → the registry flips to stopped → the settle re-poll renders fa-play again with no status residue → exactly one minimal stopAgent on the wire → the Body record returns to state: 'off' with pendingAction/controlReason null → the credential sweep runs last over ALL recorded requests.
JSDoc: the describe/test summaries state the full-cycle contract.

Acceptance Criteria

The happy-path witness drives Start → registry-running → Stop → registry-stopped on the mounted composition, each render advance sourced from a re-polled registry snapshot (never the click).
Exactly one stopAgent crosses the wire, minimal payload (registered wire method, one string param, no credential-shaped bytes).
After the settled Stop: toggle renders the Start affordance, .fm-card-control-status not visible, Body record state 'off' with no pending/reason residue.
The credential-hygiene sweep covers every request recorded across the full cycle.

Out of Scope

A rejected-Stop variant — the rejected-terminal class (honest reason render + open retry + untouched registry) is already witnessed on Start and the terminal path is verb-agnostic (fleetLifecycleIntentAdapter writes the same controlReason shape). Restart (fa-rotate) — meaningful only while running; a candidate for the matrix if the surface grows, not this slice.

#15439 (the Start leg, delivered by PR #15445) · PR #15445 review thread (the observation's origin) · #15242 (the retirement that re-homed the surface).

Live latest-open sweep: latest 20 open at 2026-07-18T11:38Z; no equivalent. A2A claim sweep: list_messages window 09:00–11:39Z — claims on #15374/#15449 (Emmy), #15251 (Clio), #15184 (Ada); none on this scope.

Decision Record impact: none.

Origin Session ID: 7157d21f-16c8-4b76-8aac-67e166deccca Retrieval Hint: "card lifecycle stop verb power cycle registry truth witness"