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	feat(testing): probe unit-suite CI wall-clock with workers 4 (#15783)
author	neo-opus-vega
state	Closed
createdAt	Jul 24, 2026, 11:14 AM
updatedAt	Jul 27, 2026, 12:04 AM
closedAt	Jul 24, 2026, 11:25 AM
mergedAt
branches	`dev` ← `agent/15783-unit-workers-probe`
url	https://github.com/neomjs/neo/pull/15784
contentTrust
projected
quarantined	0
signals	[]

Closed

neo-opus-vega commented on Jul 24, 2026, 11:14 AM

Resolves #15783

Measurement probe, one line: workers: process.env.CI ? 1 : undefined → ? 4 in test/playwright/playwright.config.unit.mjs. The CI unit job currently runs all 820 unit specs single-worker on a 4-vCPU public runner (~10-minute wall across recent runs, setup only ~25s); the line is unchanged since the config's creation (2025-10-12, #7471), while the isolation-by-construction substrate that makes parallelism plausible (per-process Chroma DB naming from the #12335 fix, UNIT_TEST_MODE isolation per ADR 0019 B4, Playwright per-file worker processes) all landed after it. This PR exists to produce the numbers the ticket's verdict needs; it merges only through the ordinary review + human gate after the verdict, and a repeatable cross-file collision closes it unmerged in favor of naming the specific isolation defect it exposed.

Evidence: L2 achieved by this PR's own CI runs (the measurement IS the artifact: the config file is a non-docs path, so the classifier runs the full unit suite at this head; wall-clock + retry telemetry recorded on #15783 against the baseline runs cited there) → L2 required (the close-target ACs are CI-measurement contracts). Residual: the ≥2-run sample + verdict comment land on the ticket after CI completes (AC 2/3).

Deltas from ticket

None substantive — the diff is exactly the ticket's prescribed single line.

Test Evidence

The probe's evidence surface is CI itself: unit job wall-clock + retries: 2 telemetry at this head, ≥2 runs, recorded on #15783 next to the single-worker baselines (9:56 and 10:36 on 2026-07-24/23).
Local single-worker runs remain unaffected (workers: undefined outside CI — unchanged).
Directly touched app/feature surfaces: None found (test-infra config only).

Post-Merge Validation

First five post-merge dev unit runs: wall-clock stays in the measured band and no new flake class appears in retry telemetry (revert is a one-line restore if it does).

Review note

Held for the #15783 verdict before any merge handoff: reviewer(s) should see the recorded measurements, not just the diff. Cross-family seat intentionally not requested until the data is on the ticket.

Authored by Vega (Claude Fable 5, Claude Code). Session 2cca0fff-6354-4036-bfdd-fc3320938015.

neo-opus-vega commented on Jul 24, 2026, 11:25 AM

Closing unmerged per the #15783 measurement protocol (branch 2): the ~2.7× wall-clock win is proven (3.6m full suite vs 9:56/10:36 single-worker baselines) but blocked by a deterministically parallel-incompatible absolute profiling budget (StoreFilterProfile, 748ms vs <400ms on 2/2 attempts, all retries). Verdict + full taxonomy: https://github.com/neomjs/neo/issues/15783#issuecomment-5068282947 — re-landing path: #15790 (parallel-compatible profiling budgets, carries the workers:4 re-land AC) + #15789 (SortZone global Neo.applyDeltas bleed, a correctness defect the probe exposed). The branch stays for the re-land reference.