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	14449
title	Codex stop-hook spec reads live prompt-context store
state	Closed
labels	bugaitestingmodel-experience
assignees	neo-gpt
createdAt	Jul 2, 2026, 9:39 AM
updatedAt	Jul 2, 2026, 9:47 AM
githubUrl	https://github.com/neomjs/neo/issues/14449
author	neo-opus-ada
commentsCount	1
parentIssue	null
subIssues	[]
subIssuesCompleted	0
subIssuesTotal	0
contentTrust
projected
quarantined	0
signals	[]
blockedBy	[]
blocking	[]
closedAt	Jul 2, 2026, 9:47 AM

Codex stop-hook spec reads live prompt-context store

Closed v13.1.0/archive-v13-1-0-chunk-8 bugaitestingmodel-experience

neo-opus-ada commented on Jul 2, 2026, 9:39 AM

Context

During PR #14439 re-review, Clio reported that test/playwright/unit/hooks/codexLaneStateStopHook.spec.mjs can read the real Codex prompt-context store when direct classification tests call classifyCodexStopPayload(fixture) without an explicit hook-store override. That makes two fixture tests depend on whether another live Codex session recently wrote codex-prompt-context.json.

This is a narrow leaf under #14420's Codex stop-hook prompt-visibility surface. It does not implement #14420 Defect B, Defect C, or the #14438 re-fire admission mechanism.

The Problem

.codex/hooks/codex-lane-state-stop.mjs computes its default prompt-context store from the module-level LOG_DIR. Direct unit tests import the module in-process, so calls that do not pass messages, last_user_message, or transcript_path can fall through to readPromptContext() and consume the real store.

That is correct for the live hook path, but wrong for pure fixture classification tests. A live prompt-context file can make a fixture that should produce promptSource=none classify as promptSource=prompt_context, changing a would-block result into an operator-dialogue allow.

The Architectural Reality

Live hook behavior must continue to use the module-level default store.
Spawned hook tests already isolate correctly by setting NEO_AI_DAEMON_DIR.
Pure classification tests need a test-owned logDir seam so they can exercise the same prompt-context fallback deliberately without reading the user's real store.
The owning files are .codex/hooks/codex-lane-state-stop.mjs and test/playwright/unit/hooks/codexLaneStateStopHook.spec.mjs.

The Fix

Add an optional logDir option to the pure classification path:

Target Surface	Source of Authority	Proposed Behavior	Fallback	Docs	Evidence
`extractPromptingText(input, {logDir})`	Existing Codex prompt-context fallback and #14420 prompt-visibility isolation	Use `logDir` when resolving `codex-prompt-context.json` for in-process classification	Omitted `logDir` preserves live default store behavior	JSDoc option entry	Focused Codex hook spec
`classifyCodexStopPayload(input, {logDir})`	Same	Forward `logDir` into prompting-text extraction	Omitted `logDir` preserves current behavior	JSDoc option entry	Pollution regression test
`codexLaneStateStopHook.spec.mjs` direct fixture tests	Unit-test isolation contract	Direct no-prompt classification tests run against an empty temp store	Spawned hook tests keep using `NEO_AI_DAEMON_DIR`	Test helper name	Focused and hook-directory unit runs

Acceptance Criteria

Direct fixture classification that expects promptSource=none uses an explicit isolated prompt-context directory.
A regression test proves a polluted prompt-context store only affects classification when the test opts into that store.
Live/spawned hook prompt-context fallback remains covered.
Focused Codex hook spec passes.
Full hook unit directory passes.

Out of Scope

#14420 Defect B operator-prompt visibility semantics.
#14420 Defect C sunset terminal reconciliation.
#14438 declining-yield re-fire admission.
Any production policy change to no-hold / stop-hook admission.

Avoided Traps

Do not remove the prompt-context fallback; live Codex UserPromptSubmit to Stop continuity needs it.
Do not make tests depend on deleting or mutating the user's real store.
Do not close #14420 from this leaf; this fixes one test isolation defect only.

#14420 #14439

Live latest-open sweep: checked latest 20 open issues at 2026-07-02T07:38:59Z; no equivalent found. Closest related ticket is #14440, but it is broader, not-code-ready, and about continuation-chain operator visibility rather than unit-test store isolation.

A2A in-flight sweep: checked latest 30 all-state mailbox messages at 2026-07-02T07:39Z; only overlapping claim was @neo-gpt's own [lane-claim] #14420 Codex stop-hook spec live-store isolation.

Origin Session ID: 8facbc96-c346-4633-9141-79a968ca1c5d Retrieval Hint: "codexLaneStateStopHook NEO_AI_DAEMON_DIR promptSource live store isolation"