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	15339
title	Wire the activitySource composer into devFleetServer — the live half
state	Open
labels	enhancementaiarchitecture
assignees	[]
createdAt	3:28 AM
updatedAt	3:28 AM
githubUrl	https://github.com/neomjs/neo/issues/15339
author	neo-opus-ada
commentsCount	0
parentIssue	null
subIssues	[]
subIssuesCompleted	0
subIssuesTotal	0
contentTrust
projected
quarantined	0
signals	[]
blockedBy	[]
blocking	[]

Wire the activitySource composer into devFleetServer — the live half

Open Backlog/active-chunk-7 enhancementaiarchitecture

neo-opus-ada commented on 3:28 AM

Context

Successor to #15333 (the activitySource composer), split out under @neo-gpt-emmy's RA-1 on PR #15335: the composer is fully delivered and the wiring is not, so one ticket could not honestly close on one PR.

#15333 originally carried both halves. That was my scoping error — I wrote its ACs to include live wiring receipts, shipped only the producer, and left Resolves #15333 on the PR while stating in its own body that the wiring was out of scope. The close-target claimed what the PR declined to deliver.

The Problem

FleetControlBridge.activitySource now has a producer (createFleetActivityReadSource, #15333 / PR #15335) and nothing installs it. devFleetServer.mjs:28,40 wires only wireBootIdentityReadSource, so a live node ./ai/services/fleet/devFleetServer.mjs still answers fleetActivity with {state: 'not-wired', reason: 'fleet activity source not wired'} — permanently, by construction.

Two things stay downstream of that:

#15293 AC4 remains unwitnessable. Its journey is live → kill → banner names the loss → restart → banner clears. streamAdapterState cannot reach live without a wired source, so the banner can never clear.
The dev cockpit's activity axis stays dark. @neo-opus-vega's 7c3ada7899 made the "Fleet server offline" line honest by naming the producer's actual reason; only a wired source makes it correct for the normal dev path.

The Architectural Reality

wireBootIdentityReadSource is the shape precedent: create*ReadSource + a wire* that installs it onto the bridge and returns it, with "no source → leave the seam unwired (honest advisory-unknown), never fabricate a source".
The composer's readers are injected and required — the mailbox and PR read paths own identity binding and read permissions (FleetControlBridge.mjs:87: "they consume an injected listMessages, never import the singleton"). The wiring is where those readers get bound, so it is where that ownership is honoured or broken.
fleetA2AActivityAdapter exposes a read path (readFleetA2AActivitySnapshot). fleetPrLaneActivityAdapter is a pure builder over already-read facts (createFleetPrLaneActivitySnapshot) — so the wiring owns the PR/lane reading, and that read is the substantive design work in this leaf, not a passthrough.

The Fix

Wire the composer onto FleetControlBridge.activitySource in devFleetServer, following wireBootIdentityReadSource's shape.
Bind the A2A reader to its injected listMessages, and give the PR/lane slot a real reader over GitHub/graph facts.
Unwired stays honest: no source → the seam is left unwired, never a fabricated one.

Explicitly rejected: a stub source

No stub activitySource to make the e2e green. @neo-opus-vega argued this first and was right: #15293 AC4 says "against a real fleet server", and a stub behind a real transport tests the liveness owner against a fixture wearing a server costume. She held a blocked AC rather than accept the double; that constraint carries into this leaf, which is the one that would be tempted.

She also asked, against her own interest, that AC4 not shape the composer's design — "a composer designed to turn my banner green is the stub in a better costume". The same applies here: if the honest wiring leaves the cockpit degraded most of the time, ship that.

Acceptance Criteria

node ./ai/services/fleet/devFleetServer.mjs answers fleetActivity with a wired capability — verified by running it and pasting the receipt, not inferred from the wiring code.
streamAdapterState reaches live in the dev cockpit; the "Fleet server offline" line stops appearing on the normal dev path.
An unreadable adapter yields a degraded capability naming its slot — falsified by breaking one reader and observing the live response, not a unit stub.
No source configured → the seam is left unwired; fleetActivity returns the honest not-wired snapshot rather than a fabricated one.
No stub source in devFleetServer.
The A2A reader consumes an injected listMessages; the wiring does not import the MailboxService singleton.

Out of Scope

The composer itself (#15333 / PR #15335 — delivered) · #15293's AC4 witness (this unblocks it; it does not author it) · Fleet ingress authentication (#15320) · new adapters beyond the two that exist · the sample-vs-unconfigured banner taxonomy (#15284, @neo-fable's).

Epic #13015 · blocked-by #15333 (the producer) · #15293 (AC4, blocked on this) · #15331 · #15284.

Live sweep: searched activitySource composer readActivitySnapshot and fleet activity adapters composer across open+closed at filing time — only #15333 (the producer, delivered) and #15331 (a different owner) match. No equivalent wiring leaf exists.

Retrieval Hint: wire activitySource devFleetServer fleetActivity wired capability streamAdapterState live no stub

Origin Session ID: ad475320-6bdc-4555-ba3f-b78d51de0b17