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.

Neo.mjs - Self-Evolving Software Organism

Frontmatter

title	fix: harden SessionService summarization payload (#9973)
author	tobiu
state	Merged
createdAt	Apr 13, 2026, 4:36 PM
updatedAt	Apr 13, 2026, 4:46 PM
closedAt	Apr 13, 2026, 4:46 PM
mergedAt	Apr 13, 2026, 4:46 PM
branches	`dev` ← `agent/9973-memory-summarization-fix`
url	https://github.com/neomjs/neo/pull/9974

Merged

tobiu commented on Apr 13, 2026, 4:36 PM

Resolution for #9973

Architectural Fix:

Refactored SessionService.mjs to aggressively parse and truncate toolsUsed configurations. This prevents autonomous LLMs from accidentally injecting massive multi-kilobyte script mutations directly into the summarization prompt.
Added 50K-character payload bounds tests in SessionSummarization.spec.mjs.
Updated openapi.yaml schema for add_memory to explicitly forbid agents from serializing JSON bodies into the logging endpoints.

Resolves #9973

tobiu commented on Apr 13, 2026, 4:44 PM

Input from Antigravity (Gemini 3.1 Pro):

✦ # PR Review Summary

Status: Approved

Thanks for putting this together! Great approach to solving the Memory Core summarization token exhaustion paradox. Isolating the LLM tool payload stringification bug at the extraction layer entirely avoids the negative ROI of aggressive graph summarization truncation. I've left the evaluation metrics below. Ready to merge!

📊 Evaluation Metrics

[ARCH_ALIGNMENT]: 100 - Perfectly aligns with the SessionService Map-Reduce extraction pipeline, treating the token limits mathematically instead of relying on prompt hacking.

[CONTENT_COMPLETENESS]: 100 - OpenAPI schema updated to formally close the trust gap, Playwright tests added demonstrating the memory bounds logic.

[EXECUTION_QUALITY]: 100 - Defensive parsing with 50-char truncation is robust against malformed LLM arrays or JSON tool dump hallucinations. Playwright offline simulation verified bounds under 15,000 chars.

[PRODUCTIVITY]: 100 - Successfully prevents the MLX 4096 n_ctx crash while keeping useful tool logging intact.

[IMPACT]: 85 - High impact for stability. Unblocks long-running autonomous RLAIF sessions from crashing the underlying memory storage during processing.

[COMPLEXITY]: 30 - Straightforward extraction refactoring and Playwright mocking.

[EFFORT_PROFILE]: Quick Win - High ROI for system stability without requiring sweeping architectural rewrites.

🕸️ Context & Graph Linking

Target Epic / Issue ID: Resolves #9973

Related Graph Nodes: SessionService, MemoryService

🧠 Graph Ingestion Notes

[KB_GAP]: Previously, agents lacked an explicit contract preventing them from storing massive payload execution parameters into the memory toolsUsed array. The OpenAPI change formally closes this systemic API trust gap.

[TOOLING_GAP]: N/A

[RETROSPECTIVE]: The "Final 3 Tuples Truncation" hypothesis was correctly rejected as a catastrophic context-loss mechanism. Protecting prompt boundaries via upstream data-stream sanitization (defensive metadata extraction) preserves full Swarm Intelligence history while respecting local LLM token limits (n_ctx safety).

📋 Required Actions

To proceed with merging, please address the following:

All checks passing and verified via unit tests. (Auto-approved)

Ship it!