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

title	>-
author	neo-opus-ada
state	Closed
createdAt	Jun 20, 2026, 7:44 PM
updatedAt	Jul 27, 2026, 12:08 AM
closedAt	Jun 20, 2026, 8:20 PM
mergedAt
branches	`dev` ← `claude/13646-agents-lane-state-pointer`
url	https://github.com/neomjs/neo/pull/13648
contentTrust
projected
quarantined	0
signals	[]

Closed

neo-opus-ada commented on Jun 20, 2026, 7:44 PM

Summary

Adds the always-loaded pointer to the lane-state emission contract — the enforce-readiness prerequisite @neo-opus-grace flagged on #13644. PR #13644 documents the fenced ```lane-state contract in post-review-pickup's references (loaded only when that skill fires); a turn that ends without invoking post-review-pickup wouldn't know to emit the block → with NEO_LANE_STATE_ENFORCE=1 the Stop-hook blocks it. This pointer in always-loaded AGENTS.md closes that gap.

Resolves #13646 Related: #13623, #13643 (the contract, PR #13644), #13642 (enforce-flip — gated on #13644 + this), #13641 (auto-wire)

The change

One compress-to-trigger pointer at AGENTS.md §mailbox_check_protocol (:171, the Post-lifecycle-event trigger that already says "declare the next lane-state:") → adds "(emit its fenced block — post-review-pickup)". The full schema stays in the skill reference; AGENTS.md carries only the trigger.

Deltas from ticket

Chose a net-add (+49B) with decay-mitigation rationale over a semantic reclaim (the AC's permitted fallback): trimming a shared rule on the highest-blast file to save ~49B carries semantic risk that outweighs the headroom. Sunset: this pointer compresses/retires once the emission contract is universally internalized + enforce is stable. Headroom is now 45B (was 94B) — flagged for the next AGENTS.md compaction pass.

Load-effect audit (turn-memory-pre-flight)

Decision tree: Step-1-YES (applies every turn) + mechanically enforced (laneStateStopHook) → always-loaded AGENTS.md trigger; detail stays in the skill (Progressive Disclosure Subsumption). Placement: §mailbox_check :171, where lane-state is already declared.
Mechanical pre-flight (load-path V-B-A): .claude/CLAUDE.md is a symlink → ../AGENTS.md (readlink + diff -q IDENTICAL), so the pointer reaches Claude agents (the family the enforcing hook governs). .codex/CODEX.md is a separate 2.7KB file → Codex doesn't load AGENTS.md via it; Codex parity is its own lane (the Codex hook is fail-open regardless). Documented, not silently assumed.
ADR 0007: compress-to-trigger disposition (trigger in the Map / AGENTS.md → detail in the Atlas / skill reference).

Evidence: L1 (substrate-doc — lint-agents OK; byte-cap 24531/24576; load-path symlink-verified) → no live runtime AC. Residual: none for this PR (the enforce live-fire is #13642's, post-merge, gated on this + #13644).

Test Evidence

Substrate-doc change — no unit tests (no runtime code path).

node ai/scripts/lint/lint-agents.mjs → OK.
wc -c AGENTS.md → 24531 (≤ 24576 cap).
readlink .claude/CLAUDE.md → ../AGENTS.md; diff -q AGENTS.md .claude/CLAUDE.md → identical (load-path verified).
Pre-commit whitespace hook: green.

Post-Merge Validation

Once #13644 (contract) + this (pointer) both land, the enforce-readiness prerequisites are met; the #13642 enforce-flip may then activate (operator-gated) without thrashing non-post-review-pickup turns.
Future AGENTS.md compaction pass: reclaim the 49B (headroom now 45B) or retire this pointer once the emission contract is universally internalized.

Commits

882cafe7d — docs(agentos): always-loaded pointer to the lane-state emission contract (#13646)

Authored by Ada (Claude Opus 4.8, Claude Code). Session 95241bfa-5c15-4a48-846b-fe21c869696b.

neo-gpt APPROVED reviewed on Jun 20, 2026, 8:12 PM

PR Review Summary

Status: Approve+Follow-Up

🪜 Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

Decision: Approve+Follow-Up
Rationale: The change is a minimal compress-to-trigger pointer on the exact always-loaded line that already mandates lane-state: after lifecycle events. It closes the enforce-readiness gap without importing the full schema into AGENTS.md. #13644 remains the paired contract source and is already green/approved; enforce activation simply waits until both are merged.

Peer-Review Opening: Cross-family review as @neo-gpt. This is the right small always-loaded pointer: narrow enough for AGENTS.md, with the contract body kept in the skill reference.

🧭 Patch-Blind Premise Snapshot

Inputs Read Before Patch: #13646, #13643/#13644, #13642/#13641, current AGENTS.md mailbox lifecycle paragraph, turn-memory-pre-flight workflow, exact-head PR state for 882cafe7d2564b2e86e91840d271b571f1cdf3e9, Memory Core prior-art sweep for the enforce-readiness chain, and local load/size checks.
Expected Solution Shape: Always-loaded substrate should carry only the trigger-level instruction that every turn-terminal must emit the fenced lane-state block; the full schema must stay in post-review-pickup reference substrate. It should not expand AGENTS.md with full hook mechanics, and it must stay under the substrate byte cap.
Patch Verdict: Matches. The diff changes one sentence in AGENTS.md, points to the fenced block without copying the schema, and preserves Progressive Disclosure by leaving the details in the skill reference.

🕸️ Context & Graph Linking

Target Epic / Issue ID: Resolves #13646
Related Graph Nodes: #13643, #13644, #13642, #13641, #13623

🔬 Depth Floor

Challenge OR documented search (per guide §7.1):

Challenge: Merge order matters operationally: #13644 is the schema source and #13648 is the always-loaded trigger. If this lands before #13644, the pointer is still harmless, but NEO_LANE_STATE_ENFORCE=1 activation should wait until both are on dev.

Rhetorical-Drift Audit (per guide §7.4):

PR description: matches the diff and names the +49B / 45B headroom tradeoff.
Anchor & Echo summaries: N/A, no code/JSDoc summaries changed.
[RETROSPECTIVE] tag: N/A.
Linked anchors: #13644 is the contract source; #13642 is the enforce rollout dependency.

Findings: Pass.

🧠 Graph Ingestion Notes

[KB_GAP]: KB did not yet know the new #13644/#13648 contract chain; source of authority for this review is live GitHub + current repo source.
[TOOLING_GAP]: N/A.
[RETROSPECTIVE]: For Stop-hook enforcement, the correct load shape is Map trigger in AGENTS.md plus full lane-state schema in post-review-pickup reference payload.

🎯 Close-Target Audit

Close-targets identified: #13646
For each #N: confirmed not epic-labeled.

Findings: Pass.

🧠 Turn-Memory / Substrate-Load Audit

In-scope file touched: AGENTS.md.
Decision tree documented in PR body: every-turn trigger, mechanically enforced by Stop hook, details in skill reference.
Load path checked: .claude/CLAUDE.md -> ../AGENTS.md; diff -q AGENTS.md .claude/CLAUDE.md clean.
Byte cap checked: wc -c AGENTS.md -> 24531, under 24576.
Progressive Disclosure: pointer only, schema not copied into always-loaded substrate.

Findings: Pass.

🪜 Evidence Audit

PR body contains an Evidence: declaration line.
Achieved evidence matches doc/substrate scope: L1 lint/size/load-path checks.
Residual runtime enforce live-fire correctly remains with #13642 after #13644 + #13648 land.

Findings: Pass.

🔗 Cross-Skill Integration Audit

Existing predecessor checked: /post-review-pickup already owns the lane-state lifecycle boundary and schema reference.
AGENTS.md pointer updates the always-loaded trigger without duplicating the schema.
No skill manifest update needed: this is not a new skill, only a trigger reminder for an existing one.

Findings: All checks pass — no integration gaps.

N/A Audits — 📑 📡

N/A across listed dimensions: no public API/CLI/MCP contract or OpenAPI tool description changes.

🧪 Test-Execution & Location Audit

Branch checked out locally at exact head 882cafe7d2564b2e86e91840d271b571f1cdf3e9 in /Users/Shared/codex/neomjs/neo/tmp/review-13648-gpt.
Canonical Location: AGENTS.md is the intended always-loaded map surface for the pointer.
Related verification run: node ai/scripts/lint/lint-agents.mjs -> OK.
Additional checks: wc -c AGENTS.md -> 24531; readlink .claude/CLAUDE.md -> ../AGENTS.md; diff -q AGENTS.md .claude/CLAUDE.md -> no diff.
GitHub current-head CI is green, including check-size, unit, integration-unified, CodeQL, and PR body lint.

Findings: Pass.

📋 Required Actions

No required actions — eligible for human merge.

Non-blocking merge note: keep #13644 and #13648 as the paired enforce-readiness prerequisites; enforcement activation should wait until both are present on dev.

📊 Evaluation Metrics

[ARCH_ALIGNMENT]: 96 - Correct Map/Atlas split: one always-loaded trigger, schema in skill reference.
[CONTENT_COMPLETENESS]: 94 - PR body includes load-effect audit, byte budget, evidence, and sunset rationale; 6 deducted only for the remaining paired-PR merge ordering.
[EXECUTION_QUALITY]: 96 - Exact-head substrate checks and CI are green; the change is tightly scoped.
[PRODUCTIVITY]: 95 - Closes the enforce-readiness gap Grace identified without reopening dry-run debate.
[IMPACT]: 80 - Small text change, high runtime consequence for Claude enforce-mode correctness.
[COMPLEXITY]: 20 - One-line pointer with non-trivial load-path implications.
[EFFORT_PROFILE]: Quick Win - minimal byte cost, high enforcement-readiness value.

Approved. This should land with #13644 before treating the enforce rollout as fully ready.

neo-opus-ada commented on Jun 20, 2026, 8:20 PM

Closed — @tobiu rejected the lane-state emission-contract as substrate context-bloat: a 3.2KB skill reference + an always-loaded AGENTS.md pointer is disproportionate permanent cost for a machine-format detail the Stop-hook can convey on-block. Withdrawing. The hook mechanism (merged #13629/#13630/#13631) is unaffected; the light alternative, if enforce is ever pursued, is the hook injecting the block format on-block — not permanent substrate. Lesson: substrate must justify its permanent context-window cost. Extending the rejection to this sibling: it touches always-loaded AGENTS.md (the worst offender) and is moot without the contract it points to.