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	8223
title	Investigate and Fix get_component_tree Timeout
state	Closed
labels	bugaiperformance
assignees	tobiu
createdAt	Dec 30, 2025, 7:48 PM
updatedAt	Dec 31, 2025, 12:47 AM
githubUrl	https://github.com/neomjs/neo/issues/8223
author	tobiu
commentsCount	2
parentIssue	8169
subIssues	[]
subIssuesCompleted	0
subIssuesTotal	0
contentTrust
projected
quarantined	0
signals	[]
blockedBy	[]
blocking	[]
closedAt	Dec 31, 2025, 12:47 AM

Investigate and Fix get_component_tree Timeout

Closed v11.18.0/archive-v11-18-0-chunk-1 bugaiperformance

tobiu commented on Dec 30, 2025, 7:48 PM

During testing of the Neural Link capabilities, the get_component_tree tool consistently timed out (30s limit) when querying an active App Worker session (2aad7032-5222-4817-90d5-8e7fa7539eca).

Tasks:

Investigate why the request is timing out.
- Is the message reaching the App Worker?
- Is the App Worker failing to serialize the tree?
- Is the response getting lost in the bridge?
Optimize the serialization performance if necessary.
Consider implementing pagination or partial tree retrieval if the full tree is too large.
Verify the fix by successfully retrieving a component tree from a live session.

tobiu Dec 30, 2025, 8:04 PM

Input from Gemini 3:

✦ I have instrumented ai/mcp/server/neural-link/services/ConnectionService.mjs with detailed logging to diagnose the timeout issue.

Changes:

Redirected spawnBridge stdout/stderr to a local ./bridge.log file.

Added logger.info to call to track outgoing requests.

Added logger.info to handleAppMessage to track incoming responses from the App Worker.

Next Steps: Restart the MCP server (and Gemini CLI) to load the instrumented code, then reproduce the timeout and analyze ./bridge.log.

tobiu Dec 31, 2025, 12:47 AM

Input from Antigravity:

◆ Resolution Report

We have successfully diagnosed and resolved the get_component_tree timeout and correctness issues.

1. Root Cause Analysis

Timeout (30s): The timeout was not caused by slow component serialization, but by an unhandled edge case in Bridge.mjs. When an App Worker reloaded and obtained a new session ID, the Agent continued sending requests to the old (stale) ID. The Bridge dropped these messages silently, causing the Agent to hang until the 30s timeout.

Incorrect Tree Data: The original implementation used Neo.manager.Component.getChildren(), which relied on VNodeUtil.getChildIds(). This method stops at component boundaries, failing to return the deep component tree.

2. Fixes Implemented

A. Logic & Correctness (ComponentService.mjs)

We replaced the retrieval method with getChildComponents, which correctly traverses the full hierarchy.
// src/ai/client/ComponentService.mjs
- const children = Neo.manager.Component.getChildren(component);
+ const children = Neo.manager.Component.getChildComponents(component);
Status: Verified 1-line change cleanliness.

B. Transport Reliability (Bridge.mjs)

Updated handleAgentMessage to send an immediate JSON-RPC error response if the target App ID is not found.
// ai/mcp/server/neural-link/Bridge.mjs
if (agentWs && payload.message && payload.message.id) {
    agentWs.send(JSON.stringify({
        type: 'app_message',
        appWorkerId: payload.target,
        message: {
            id: payload.message.id,
            error: { code: -32000, message: `Target App [${payload.target}] not found` }
        }
    }));
}
Status: Verified. Eliminates the 30s hang, failing fast instead.

3. Process Improvements

Established Anti-Reformatting Protocol in .agent/ANTIGRAVITY_RULES.md and linked it to the agent context.

Enforced sed usage for surgical edits to prevent "formatting noise" in diffs.

The issue is resolved and the codebase is clean.