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	fix: Add cwd to PullRequestService execAsync calls (#9895)
author	tobiu
state	Merged
createdAt	Apr 11, 2026, 10:15 PM
updatedAt	Apr 11, 2026, 10:19 PM
closedAt	Apr 11, 2026, 10:19 PM
mergedAt	Apr 11, 2026, 10:19 PM
branches	`dev` ← `agent/9895-pr-service-cwd-fix`
url	https://github.com/neomjs/neo/pull/9896

Merged

tobiu commented on Apr 11, 2026, 10:15 PM

Summary

Resolves #9895

Changes

PullRequestService.mjs — Two-line fix

Added {cwd: aiConfig.projectRoot} to both execAsync calls:

- const {stdout} = await execAsync(`gh pr checkout ${prNumber}`);
+ const {stdout} = await execAsync(`gh pr checkout ${prNumber}`, {cwd: aiConfig.projectRoot});

- const {stdout} = await execAsync(`gh pr diff ${prNumber}`);
+ const {stdout} = await execAsync(`gh pr diff ${prNumber}`, {cwd: aiConfig.projectRoot});

Root Cause

PullRequestService was the only service that didn't pass {cwd} to execAsync. SyncService passes it on all 5 of its calls. When the MCP server process is spawned via stdio transport (e.g., from Antigravity/Gemini CLI), the process CWD can be / rather than the git repo root. gh pr diff and gh pr checkout require being inside a git repository to resolve the PR context.

Verification

Verified live after applying the fix and restarting the MCP server:

Before: get_pull_request_diff(9894) → GH_CLI_ERROR, exit code 1
After: get_pull_request_diff(9894) → ✅ Full diff content returned

Scope

1 file, 2 insertions, 2 deletions. Zero architectural risk.

tobiu commented on Apr 11, 2026, 10:17 PM

Input from Claude Opus 4.6 (Antigravity):

❋ # PR Review Summary

Status: Approved

Two-line bug fix with live verification. The PR body is a textbook Fat Ticket — root cause analysis, before/after verification, scope boundary. The fix itself was validated by using the get_pull_request_diff MCP tool to review this very PR, which is a satisfying closed loop.

📊 Evaluation Metrics

[ARCH_ALIGNMENT]: 100 - Aligns PullRequestService with the existing SyncService pattern. Uses aiConfig.projectRoot which is already the canonical repo root resolved in config.mjs line 9.

[CONTENT_COMPLETENESS]: 95 - PR body documents root cause, verification steps, and scope. The existing JSDoc on getPullRequestDiff and checkoutPullRequest didn't need changes since the cwd option is an internal implementation detail, not API surface. Minor deduction: no inline code comment explaining why cwd is needed (though the git log will carry the Fat Ticket context via squash merge).

[EXECUTION_QUALITY]: 100 - Fix is minimal and correct. Verified live with MCP server restart. No side effects — aiConfig.projectRoot is already imported and used throughout the server. Both get_pull_request_diff and get_conversation now work (the latter was already working via GraphQL, not gh CLI).

[PRODUCTIVITY]: 100 - Bug discovered, ticketed, fixed, and verified in-session. The MCP server restart verification loop was an efficient strategy.

[IMPACT]: 60 - Unblocks get_pull_request_diff and checkout_pull_request MCP tools for all agents operating in headless/stdio environments. Impact is moderate because agents could fall back to gh pr diff via shell (as we did during the #9894 review).

[COMPLEXITY]: 5 - Two lines changed. No logic, no branching, no tests needed.

[EFFORT_PROFILE]: Quick Win - Textbook quick win. High ROI (unblocks MCP tooling) with near-zero complexity.

🕸️ Context & Graph Linking

Target Epic / Issue ID: Resolves #9895

Related Graph Nodes: #9894 (PR where the bug was discovered during review), PullRequestService.mjs, SyncService.mjs (reference implementation for the {cwd} pattern), aiConfig.projectRoot

🧠 Graph Ingestion Notes

[KB_GAP]: None. The bug was an implementation oversight, not a knowledge gap.

[TOOLING_GAP]: The MCP server's get_pull_request_diff tool silently failed with a generic GH_CLI_ERROR without indicating that the root cause was a CWD mismatch. Future improvement: the error handler could detect CWD-related failures and surface a more diagnostic message (e.g., "gh pr diff requires being in a git repository — ensure cwd is set").

[RETROSPECTIVE]: This fix validates the "restart-and-verify" workflow for MCP server changes. The user's suggestion to restart Antigravity mid-session to reload MCP servers was the correct verification strategy — it proved the fix works in the actual headless environment, not just from a developer shell. This pattern should be the standard for MCP server bug fixes.

📋 Required Actions

None — approved for merge.