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	19033
title	The PR template hands every contributor `Closes #`, including for a partial slice
state	Closed
labels	bugdocumentationcontributor-experienceai
assignees	neo-opus-ada
createdAt	Sep 21, 2026, 2:07 PM
updatedAt	Sep 21, 2026, 4:52 PM
githubUrl	https://github.com/neomjs/neo/issues/19033
author	neo-opus-grace
commentsCount	1
parentIssue	null
subIssues	[]
subIssuesCompleted	0
subIssuesTotal	0
contentTrust
projected
quarantined	0
signals	[]
blockedBy	[]
blocking	[]
closedAt	Sep 21, 2026, 4:52 PM

The PR template hands every contributor `Closes #`, including for a partial slice

Closed v13.2.0/archive-v13-2-0-chunk-22 bugdocumentationcontributor-experienceai

neo-opus-grace commented on Sep 21, 2026, 2:07 PM

Context

Found while reviewing PR #19032, the first slice of #19026 by a first-time contributor (@wofiporia). Their PR delivers the data/ directory — 7 of the ticket's 35 spec files — and their own body enumerates the thirteen directories still to do. It opens with Closes #19026, which will close the ticket on merge with 28 files undone.

They did exactly what we told them to. .github/PULL_REQUEST_TEMPLATE.md:10 is the literal line Closes #, under the heading "Does this PR resolve an issue? (Required)". There is no other option offered and no note about partial delivery.

Sweep attestations (2026-09-21T12:0xZ):

(i) Live latest-open: latest 20 open issues read; no equivalent. #18985 (@neo-opus-ada, the Hacktoberfest contributor door) is the nearest neighbour and does not cover this — none of its seven ACs mentions the PR template, and PULL_REQUEST_TEMPLATE does not appear in its body. AC-5 covers the reviewer-side "review, label and spam path", not the author-side form.
(ii) A2A in-flight: mailbox scanned over the current window; active claims are github-content-sync #8/#9 (@neo-opus-vega), Brain #103/#396 (@neo-gpt), #19027 (mine). No overlap.
(iii) Memory Core: query_raw_memories on the symptom surfaced a dense prior-art cluster — @neo-opus-ada's #12570 close-target over-claim corrected to a narrow Resolves #12574 + Refs #12536, and a three-incident session on magic-keyword parsing (\b(KEYWORD)\s+#N matches through qualifiers, so "Resolves part of #N" and even "does NOT close #N" both fire). Every one of those is agent-side discipline. None of it reaches a contributor, because none of it is in the template they are handed.
(iv) Own-assignment: #19027 only; unrelated surface.
(1b) Meta-skill: N/A — no .agents/skills/ change proposed.
(1c) Structure map: N/A — no .mjs created or relocated.

The Problem

dev is this repository's default branch, so a closing keyword fires on merge. That is not theoretical here: PR #19023 merged to dev at 11:08Z today and auto-closed #19007. GitHub already registers the link on #19032 — closingIssuesReferences: [19026].

Nothing warns. baseline / PR body passed on #19032. That check enforces the Resolves #N grammar for agent/ai PRs; a non-agent author writing the template's own Closes #N is outside its scope, so the lint is green and the close-target is still wrong.

The failure is specific to the tickets we most want newcomers in. A good first issue sized as "one directory at a time" — #19026 is exactly that shape, and #19028 and #19030 are adjacent — invites exactly one slice per PR. The template then converts each slice into a full close. The first contributor to take a slice silently ends the ticket for everyone behind them, and the damage is invisible at review time unless a reviewer thinks to check the ratio of delivered files to ticket scope.

The Architectural Reality

.github/PULL_REQUEST_TEMPLATE.md:10 — Closes #, under a heading marked (Required), with two HTML comments below it that address only the case of no issue, never the case of a partial one.
CONTRIBUTING.md — documents setup and the review path; says nothing about close-target scope.
The agent-side rule lives in .agents/skills/pull-request/references/pull-request-workflow.md §9.1 (Resolves only for the fully-delivered leaf; Closes/Fixes forbidden; broad parents get Refs) and is enforced by baseline / PR body — for ai-labelled PRs. The two audiences have opposite instructions and only one of them is enforced.

The Fix

Give the template a second, equally prominent option, and one sentence explaining when to pick it. Keep it shorter than the rule it replaces — a contributor reads this form once, under friction.

Shape (wording to settle in the PR, not prescribed here):

**Does this PR resolve an issue?** (Required)

Closes #        <!-- this PR finishes the issue -->
Refs #          <!-- this PR is one slice; the issue stays open for the rest -->

Two follow-on questions the implementation must answer rather than assume:

Does CONTRIBUTING.md need the same sentence, or does the template carry it alone? Prefer one site.
Should baseline / PR body grow a warning — not a failure — when a non-agent PR's Closes target has visibly more scope than the diff? Probably out of scope for this leaf, and worth a follow-up only if the template change proves insufficient.

Prescription checked: .github/PULL_REQUEST_TEMPLATE.md owns the concern. No better owner — CONTRIBUTING.md is read before the work, the template at the moment the decision is made.

Contract Ledger Matrix

Target surface	Source of authority	Proposed behaviour	Fallback	Docs	Evidence
`.github/PULL_REQUEST_TEMPLATE.md` issue-reference block	this ticket	offers both a closing and a non-closing form, with one line on when each applies	contributor writes neither; the existing "PRs without issue references may be closed" note already covers that	the template itself	a partial-slice PR opened after the change carries `Refs` and `closingIssuesReferences` is empty

Decision Record impact: none — no ADR governs contributor-facing templates.

Acceptance Criteria

.github/PULL_REQUEST_TEMPLATE.md offers a non-closing reference alongside Closes #, with one sentence distinguishing them, and the block is no longer than the current one plus two lines.
The wording is verified against a real case: #19032's situation, stated back in the contributor's own terms, is unambiguously answerable from the template alone.
The guidance lives in exactly one place — if CONTRIBUTING.md gains a sentence, the template links to it rather than repeating it.
No change to baseline / PR body's pass/fail behaviour in this ticket.

Out of Scope

The agent-side Resolves-only rule and its lint; those are correct and enforced.
A lint that fails a non-agent Closes overclaim. Named as a possible follow-up above; a form that reads correctly is the cheaper fix to try first.
Re-scoping #19026 or any other multi-slice ticket.
#18985's contributor-door work. This is one form field; that epic is the door.

Avoided Traps

Treating this as the contributor's mistake. It is not; they followed the only instruction given. A ticket framed as "reviewers should catch bad close targets" would put the cost on every future review instead of removing it once.
Adding a lint before fixing the form. The form says Closes # and nothing else. Enforcing against a rule we never wrote down is how the agent-side discipline stayed invisible to contributors in the first place.
Folding it into #18985. That epic's ACs are the door, the issues, CONTRIBUTING, and the merge window; a template field is a leaf, and it can land now rather than at that epic's pace. Related, not absorbed.

#19032 — the PR that surfaced it; review 5266388586 carries the measurement.
#19026 — the multi-slice ticket at risk, plus #19028 / #19030 of the same shape.
#18985 — the contributor door (@neo-opus-ada). Related; this leaf is narrower and independently landable.
#19023 / #19007 — the live control proving a dev merge auto-closes.

Handoff Retrieval Hints

query_raw_memories: "PULL_REQUEST_TEMPLATE Closes partial slice auto-close contributor"
Reproduce the hazard: gh pr view 19032 --json closingIssuesReferences against the file count in its own body table.
Prior art that shaped the framing: @neo-opus-ada's #12570 → #12574 narrow-close correction, and her three-incident magic-keyword session (qualifiers do not neutralise the parser).

Origin Session ID: 24396022-ee78-42d8-8935-0c860eb992bc

tobiu referenced in commit 76b14cc - "fix(docs): the PR template offers a non-closing reference beside Closes (#19033) (#19035) on Sep 21, 2026, 4:52 PM

tobiu closed this issue on Sep 21, 2026, 4:52 PM