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	9903
title	RLAIF Trajectory Curation & Whitebox E2E Pre-Flight
state	Closed
labels	enhancementai
assignees	tobiu
createdAt	Apr 12, 2026, 12:10 PM
updatedAt	Apr 14, 2026, 7:49 AM
githubUrl	https://github.com/neomjs/neo/issues/9903
author	tobiu
commentsCount	0
parentIssue	null
subIssues	[]
subIssuesCompleted	0
subIssuesTotal	0
blockedBy	[x] 9913 fix(ai): Implement JSON recovery/repair in DreamService Tri-Vector Synthesis
blocking	[]
closedAt	Apr 14, 2026, 12:25 AM

RLAIF Trajectory Curation & Whitebox E2E Pre-Flight

Closed v13.0.0/archive-v13-0-0-chunk-3 enhancementai

tobiu commented on Apr 12, 2026, 12:10 PM

RLAIF Trajectory Curation & Whitebox E2E Pre-Flight (formerly `executeNLActionDigest`)

[!CAUTION] ARCHITECTURAL PIVOT: This ticket was originally scoped to implement executeNLActionDigest inside the DreamService.mjs daemon to autonomously generate .spec.mjs Playwright scaffolding. That approach was identified as a critical anti-pattern producing fragile, low-ROI test noise.

The architecture has officially pivoted from "Test Generation" to "RLAIF Trajectory Curation".

Context & Motivation

The nl_action_log captured during Frontier agent (e.g., Gemini 3.1 Pro / Claude Opus 4.6) interactions via the Neural Link represents highly valuable "Demonstrations of Intelligence". Instead of stuffing these sequences blindly into Playwright test scaffolding, we will stockpile these successful component state interactions as synthetic training data to feed future SLM fine-tuning pipelines (SFT/DPO for Gemma 4).

Architectural Requirements

The Telemetry Script (No MCP Bloat): Instead of a daemon, the telemetry must be queried interactively via node ai/scripts/analyzeNlTelemetry.mjs <sessionId>. This protects the 100-tool MCP limit while allowing Frontier models to compress their own Action Logs.
RLAIF Data Persistence: Synthesized interaction sequences must be formatted functionally into .jsonl and permanently stockpiled into .neo-ai-data/datasets/rlaif/trajectories.jsonl.
Whitebox E2E Pre-Flight: Establish the .agent/skills/whitebox-e2e Progressive Disclosure skill to mandate that Frontier agents always introspect the Neural Link VDOM state before writing Playwright selectors.

References

Origin Session ID (Pivot): f191cbb2-133b-43ac-bd7a-a2e85ea1fd95 (Agents: use get_session_memories to pull context on why test generation was abandoned).
Related PR: #9902

tobiu added the enhancement label on Apr 12, 2026, 12:10 PM

tobiu added the ai label on Apr 12, 2026, 12:10 PM

tobiu marked this issue as being blocked by #9913 on Apr 12, 2026, 1:40 PM

tobiu changed title from Implement executeNLActionDigest in DreamService for Neural Link Data Synthesis to RLAIF Trajectory Curation & Whitebox E2E Pre-Flight on Apr 13, 2026, 11:35 PM

tobiu assigned to @tobiu on Apr 13, 2026, 11:37 PM

tobiu referenced in commit b53af5b - "feat(ai): Implement Neural Link telemetry extraction script and unit test (#9903)" on Apr 14, 2026, 12:13 AM

tobiu referenced in commit ccacb6a - "docs: Scaffold whitebox-e2e Progressive Disclosure skill (#9903)" on Apr 14, 2026, 12:13 AM

tobiu referenced in commit afebeae - "docs: finalized RLAIF documentation mappings and fixed namespace debt (#9903)" on Apr 14, 2026, 12:13 AM

tobiu cross-referenced by PR #9990 on Apr 14, 2026, 12:13 AM

tobiu referenced in commit ea8a0a0 - "test(ai): Fix CodeQL uncontrolled absolute path execution in telemetry tests (#9903)" on Apr 14, 2026, 12:18 AM

tobiu referenced in commit f3b68d7 - "feat/docs: RLAIF Trajectory Curation & Whitebox E2E Pre-Flight (#9903) (#9990) on Apr 14, 2026, 12:25 AM

tobiu closed this issue on Apr 14, 2026, 12:25 AM

tobiu cross-referenced by #10132 on Apr 20, 2026, 9:14 PM

tobiu cross-referenced by PR #10133 on Apr 20, 2026, 9:21 PM

tobiu referenced in commit d289889 - "chore(ai): retire manage_database_backup MCP tools (#10132) (#10133) on Apr 20, 2026, 9:41 PM