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	13413
title	First-widget evidence pane + deterministic grid render-together
state	Closed
labels	enhancementaiarchitecture
assignees	neo-opus-ada
createdAt	Jun 16, 2026, 9:34 AM
updatedAt	Jun 16, 2026, 10:32 AM
githubUrl	https://github.com/neomjs/neo/issues/13413
author	neo-opus-ada
commentsCount	0
parentIssue	13349
subIssues	[]
subIssuesCompleted	0
subIssuesTotal	0
blockedBy	[]
blocking	[]
closedAt	Jun 16, 2026, 10:32 AM

First-widget evidence pane + deterministic grid render-together

Closed v13.1.0/archive-v13-1-0-chunk-3 enhancementaiarchitecture

neo-opus-ada commented on Jun 16, 2026, 9:34 AM

Context

Split from #13355 per the #13409 cross-family review (close-target remediation, pr-review/references/close-target-remediation.md). #13355 was re-scoped to consume the #13362/#13363 Neural-Link-created grid path; PR #13409 delivered the deterministic half — the safe evidence-pane UI + a local blueprint-driven grid render-together — but NOT the NL-create-path provenance integration. This leaf is the fully-delivered close-target for that deterministic slice; #13355 stays open for the NL-created-grid provenance residual.

Release classification: post-release Agent Harness product line → boardless.

The Problem

H2 needs the visible evidence/provenance surface (request + response + accepted-blueprint metadata) rendered safely, alongside a live grid, as a deterministic + smoke-testable slice — independent of the harder NL-create-path provenance integration. Delivering this slice as its own leaf lets it close truthfully while the NL-provenance integration (#13355's residual) is designed separately, rather than overclaiming #13355's full re-scoped intent.

The Architectural Reality

apps/agentos/childapps/widget/util/blueprintEvidence.mjs — the safe allowlist projection (scalar metadata only, fail-closed).
apps/agentos/childapps/widget/view/EvidencePane.mjs — request/response/metadata via safe vdom text nodes (no html/innerHTML).
apps/agentos/childapps/widget/view/Viewport.mjs — mounts the EvidencePane + a Neo.grid.Container from ONE shared deterministic blueprint.
Tests: blueprintEvidence.spec.mjs + view/EvidencePane.spec.mjs (safe-render source assertion) + test/playwright/e2e/FirstWidgetEvidencePane.spec.mjs (render-together).
Distinct from #13362/#13363 (the NL-create-path proof, which targets examples/grid/bigData).

The Fix

Delivered by PR #13409 (re-pointed to this leaf):

The safe blueprint-evidence projection (allowlist, fail-closed).
The EvidencePane (safe vdom text, fail-closed on invalid blueprint).
The Viewport wiring: EvidencePane + a deterministic-blueprint grid (render-together), one shared blueprint = the provenance link.
Unit coverage (valid/invalid/safe-render source) + a render-together e2e.

Contract Ledger Matrix

Target Surface	Source of Authority	Proposed Behavior	Fallback	Docs	Evidence
Evidence pane (request/response/metadata)	#13349, ADR 0020, #13018	Safe vdom text only; deterministic request + response + accepted-blueprint metadata	invalid blueprint → bounded rejected-state text	JSDoc/@summary	unit + e2e (shipped green)
Blueprint-evidence projection	#13355 blueprint contract	scalar metadata only (schema/title/col-count/row-count); allowlist	unknown keys/malformed → fail closed	JSDoc/@summary	unit, 4 tests (shipped)
Deterministic grid render-together	ADR 0020	one shared blueprint drives both the evidence pane + a live grid	n/a (deterministic)	JSDoc	render-together e2e (shipped)

Decision Record impact

Aligned with ADR 0020. Split from #13355 (does NOT supersede it — #13355 retains the NL-create-path provenance scope).

Acceptance Criteria

Evidence pane renders deterministic request + response + accepted-blueprint metadata via safe vdom text (no unsafe HTML).
Invalid evidence fails closed to a bounded visible state.
A live grid renders alongside, driven by the SAME deterministic blueprint (render-together).
Unit coverage: valid/invalid evidence + safe-render source assertion.
Render-together e2e: evidence pane + grid render together.

(all delivered by PR #13409)

Out of Scope

The NL-create-path grid provenance integration (tying the evidence pane to a grid created via the #13362/#13363 create_component seam) — stays on #13355.
LLM / streaming / persistence / git-write / Electron.

Parent epic: #13349
Split from: #13355 (retains the NL-provenance residual)
Delivered by: PR #13409
Distinct foundation: #13362 / #13363 (NL-create-path proof)

Origin Session ID: 47b6dbc0-7673-4ad3-a9f5-bef3b606c56b

tobiu closed this issue on Jun 16, 2026, 10:32 AM

tobiu referenced in commit c8255b1 - "feat(agentos): first-widget evidence pane + live grid render-together (#13355) (#13409) on Jun 16, 2026, 10:32 AM