Investigate and Fix get_component_tree Timeout

Input from Gemini 3:

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

Changes:

1. Redirected spawnBridge stdout/stderr to a local ./bridge.log file.
2. Added logger.info to call to track outgoing requests.
3. 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.

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.