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	chore(ai): add self-review detection to pr-review skill (#10054)
author	tobiu
state	Merged
createdAt	Apr 17, 2026, 8:35 PM
updatedAt	Apr 17, 2026, 8:38 PM
closedAt	Apr 17, 2026, 8:38 PM
mergedAt	Apr 17, 2026, 8:38 PM
branches	`dev` ← `agent/10054-pr-review-self-detection`
url	https://github.com/neomjs/neo/pull/10055

Merged

tobiu commented on Apr 17, 2026, 8:35 PM

Summary

Adds a self-review detection heuristic to the pr-review skill so agents use first-person clinical tone when reviewing their own PRs instead of third-person congratulatory language.

Problem

When an agent authors a PR and then reviews it in the same session, the review reads as self-congratulation:

"Thanks for putting this together! Great approach... LGTM!"

This sounds off when the reviewer and the author are the same entity.

Detection Heuristic

The detection uses the Memory Core session ID as the discriminator:

After get_conversation, extract ticket number from PR body (Resolves #N)
Query query_raw_memories(query: '#N') scoped to current session ID
Match → self-review mode (first-person, clinical)
No match → peer-review mode (third-person, constructive)

This correctly handles all four author/reviewer combinations:

Author	Reviewer	Session Match	Mode
Agent (session A)	Same agent (session A)	✅	Self-review
Agent (session A)	Different agent (session B)	❌	Peer-review
Human	Agent	❌	Peer-review
External contributor	Agent	❌	Peer-review

Changes

pr-review-guide.md

Section 1: New Self-Review (same session) philosophy bullet
Section 2: New step 2 Self-Review Detection with session-based authorship check
Renumbered existing steps 2→3, 3→4, 4→5

pr-review-template.md

Dual opening block: peer-review (friendly) vs self-review (clinical)
Usage note: "Use exactly one opening based on detection result"

Resolves #10054

tobiu commented on Apr 17, 2026, 8:36 PM

Input from Claude Opus 4.6 (Antigravity):

❋ # PR Review Summary

Status: Approved

Self-Review Opening: Self-review of #10054. This implementation chose a session-based Memory Core heuristic over a simple GitHub login comparison because the login is shared infrastructure (human and agent both operate as tobiu). The session ID is the true identity discriminator.

📊 Evaluation Metrics

[ARCH_ALIGNMENT]: 90 - Uses the existing Memory Core query infrastructure (query_raw_memories) rather than inventing new detection mechanisms. No new tools or code needed.

[CONTENT_COMPLETENESS]: 90 - Both the guide and template are updated. The detection heuristic is documented with clear step-by-step instructions.

[EXECUTION_QUALITY]: 85 - Clean diff. One minor concern: the detection relies on Resolves #N being present in the PR body. If an agent omits the ticket reference, detection falls back to peer-review mode (safe default, but worth noting).

[PRODUCTIVITY]: 100 - Matches ticket scope exactly.

[IMPACT]: 50 - Improves review quality but is a tone/style enhancement, not a functional change.

[COMPLEXITY]: 15 - Two markdown file edits.

[EFFORT_PROFILE]: Quick Win - Minimal complexity for a meaningful reduction in uncanny-valley self-congratulation.

🕸️ Context & Graph Linking

Target Epic / Issue ID: Resolves #10054

Related Graph Nodes: pr-review skill, Memory Core query_raw_memories, ticket-intake skill (#10051/#10053)

🧠 Graph Ingestion Notes

[KB_GAP]: N/A

[TOOLING_GAP]: N/A

[RETROSPECTIVE]: The session-based identity pattern (Memory Core session ID as the agent fingerprint) has broader applicability beyond PR reviews. Any workflow that needs "did I do this?" authorship detection can reuse this heuristic. This is the first time we've formally distinguished agent identity from GitHub account identity in a protocol document.

📋 Required Actions

None. Ready to merge.