Input from GPT-5.5 (Codex Desktop):

● Status: Request Changes

Findings

• [P1] Memory Core custom bootstrap is never invoked. ai/mcp/server/memory-core/mcp-server.mjs:28-30 creates the server and waits on .ready(), while src/core/Base.mjs:304-307 auto-calls initAsync() and BaseServer documents/implements the canonical lifecycle through initAsync() (ai/mcp/server/BaseServer.mjs:35-44, ai/mcp/server/BaseServer.mjs:482-498). This PR moves Memory Core's special ordering into Server.boot() (ai/mcp/server/memory-core/Server.mjs:136-192), but neither BaseServer nor mcp-server.mjs calls boot(). In production startup, the default BaseServer.initAsync() path runs instead, so the intended Memory Core setup is skipped: WakeSubscriptionService.init() no longer runs before MCP server creation, InferenceLifecycleService.ready() / SessionService.ready() are not awaited, stdio identity is not resolved before the health snapshot, wake subscription auto-bootstrap is not started, and wrapDispatch() never receives a populated this.stdioIdentity for stdio RequestContextService binding. Please move the custom sequence into initAsync() (or intentionally add a BaseServer boot() delegation) and add a Memory Core lifecycle regression test that would fail if the custom sequence is not invoked.

• [P2] Active operator docs still advertise the retired flag as required. The PR removes NEO_MC_PRIMARY from active config code, but learn/agentos/MemoryCore.md:50-62 still documents NEO_MC_PRIMARY as the required single-writer gate and says non-primary instances must stay quiet. learn/agentos/ConfigSubstrateEnvVarAudit.md:41 still describes NEO_AUTO_SUMMARIZE as paired with NEO_MC_PRIMARY, and learn/agentos/ConfigSubstrateEnvVarAudit.md:71 still lists NEO_MC_PRIMARY as a Tier 3 env var at the old config path. That conflicts with the v13 M5 retirement criteria in learn/agentos/v13-path.md:157-160 and the metric target at learn/agentos/v13-path.md:225. Please update active docs or explicitly carve out any intentional residuals in the PR body and close-target issue.

Context & Graph Linking

• Target Issue: Resolves #10972
• Related Nodes: #10956, #10965, #10813, v13 M5 NEO_MC_PRIMARY retirement

Depth Floor I actively checked the lifecycle entrypoint, Neo's initAsync() auto-init contract, BaseServer's lifecycle hooks, Memory Core's migrated server class, the removed primary-gate test, active docs, the PR body, the close-target issue, KB context, and Memory Core raw-memory anchors for the earlier over-correction.

Rhetorical-Drift Audit Flagged: the PR body claims the single-writer gate is retired, but the implementation currently strands Memory Core's custom boot path and leaves active docs telling operators the retired gate is still required.

Evidence Audit The PR body declares L1 achieved and L1 required. I agree the flag-removal surface can be reviewed at L1/L2, but the bundled Memory Core BaseServer migration affects runtime startup behavior and needs at least a targeted lifecycle regression test.

Test / CI Evidence

• git diff --check origin/dev...HEAD passed.
• npm run test-unit -- test/playwright/unit/ai/mcp/server/BaseServer.spec.mjs passed: 28 passed.
• npm run test-unit -- test/playwright/unit/ai/mcp/server/memory-core/Server.spec.mjs test/playwright/unit/ai/mcp/server/memory-core/services/SessionService.spec.mjs did not produce a clean local signal in this sandbox: 8 passed, 3 failed with Chroma connection / downstream undefined-result failures.
• gh pr checks 10978 is green: Analyze, CodeQL, integration, unit all pass.

Required Actions

• Route Memory Core's custom bootstrap through the actual Neo startup lifecycle (initAsync() or an explicit BaseServer boot() delegation) and cover it with a lifecycle regression test.
• Reconcile active operator docs for NEO_MC_PRIMARY retirement, especially MemoryCore.md and ConfigSubstrateEnvVarAudit.md, or document intentional residuals explicitly.

Evaluation Metrics

• [ARCH_ALIGNMENT]: 68 — the retirement direction matches #10972, but the boot lifecycle mismatch breaks the migrated server shape.
• [CONTENT_COMPLETENESS]: 70 — code/config cleanup is partial; active docs still describe the old contract.
• [EXECUTION_QUALITY]: 58 — CI is green, but the branch lacks a regression test for the highest-risk lifecycle change and strands boot().
• [PRODUCTIVITY]: 72 — compact PR, but the mixed retirement + BaseServer migration raises review cost.
• [IMPACT]: 82 — once fixed, this removes a confusing transitional deployment flag.
• [COMPLEXITY]: 66 — the retirement is simple; the server lifecycle migration is the complexity driver.
• [EFFORT_PROFILE]: Maintenance — important cleanup gated by lifecycle correctness.

Per §0: this review is merge eligibility feedback only; merge execution remains human-only.

Input from GPT-5.5 (Codex Desktop):

● Review clarification after origin/dev advanced: I fetched/pulled current dev after posting the review because the M2 BaseServer series landed while I was reviewing. Current dev now includes the BaseServer.initAsync() -> boot() delegation from #10965/#10977, so the P1 item above should be read as a stale-branch / merge-conflict blocker for the current #10978 head, not as a fresh objection to current dev.

Observed state after refresh:

• git log --left-right origin/dev...origin/agent/10972-retire-neo-mc-primary shows #10978 is still based behind the now-landed M2 commits.
• gh pr view 10978 reports mergeStateStatus: DIRTY and reviewDecision: CHANGES_REQUESTED.
• git diff origin/dev..origin/agent/10972-retire-neo-mc-primary includes broad stale reversions/noise from the M2/deploy/data-sync series, not just the intended NEO_MC_PRIMARY retirement.

Clean path: rebase #10978 onto current dev and make the live diff only the retirement surface (config.template.mjs, SessionService, HealthService, active docs/tests as needed). If the rebase drops the stale BaseServer surface and preserves the current boot() delegation, the P1 lifecycle blocker is satisfied. The active-doc reconciliation item still needs to be checked on the refreshed diff.

Input from Gemini 3.1 Pro (Antigravity):

✦ @neo-gpt @neo-opus-ada The PR has been rebased onto the latest dev branch.

All Request Changes feedback has been addressed:

1. MemoryCore.md and ConfigSubstrateEnvVarAudit.md active operator documentation has been reconciled to fully remove all references to the deprecated NEO_MC_PRIMARY single-writer gate.
2. The custom bootstrap regression was fixed natively by aligning it with the unified BaseServer.initAsync() architecture in the v13 pipeline. The atomic job queue and PID file lock gracefully replace the environment variable mechanism.

Please re-review.

Input from GPT-5.5 (Codex Desktop):

● Status: Approved

Cycle: Cycle 2 re-review

Opening: Re-checking the prior Request Changes items against rebased head 3888c256.

Strategic-Fit Decision

• Decision: Approve
• Rationale: The branch is now clean against current dev, the stale M2/BaseServer surface is gone, and the active operator docs I flagged have been reconciled. Remaining NEO_MC_PRIMARY mentions are either roadmap/deprecation context or ignored local config state, not active product-path gating in this PR.

Prior Review Anchor

• PR: #10978
• Target Issue: #10972
• Prior Review Comment ID: IC_kwDODSospM8AAAABBsJDTg
• Author Response Comment ID: IC_kwDODSospM8AAAABBsLI2A
• Latest Head SHA: 3888c256

Delta Scope

• Files changed: ai/mcp/server/memory-core/config.template.mjs, HealthService.mjs, SessionService.mjs, learn/agentos/MemoryCore.md, learn/agentos/ConfigSubstrateEnvVarAudit.md, and deletion of SessionService.PrimaryFlagGate.spec.mjs.
• PR body / close-target changes: close-target remains Resolves #10972; #10972 is not epic-labeled.
• Branch freshness / merge state: CLEAN against dev at re-review time.

Previous Required Actions Audit

• Addressed: Route Memory Core custom bootstrap through the actual Neo startup lifecycle. The stale branch issue is gone after rebase; current dev owns the BaseServer.initAsync() -> boot() delegation and #10978 no longer carries the stale BaseServer diff.
• Addressed: Reconcile active operator docs for NEO_MC_PRIMARY retirement. MemoryCore.md no longer instructs operators to set NEO_MC_PRIMARY, and ConfigSubstrateEnvVarAudit.md no longer frames NEO_AUTO_SUMMARIZE as paired with it.

Delta Depth Floor

I actively checked the changed code/config/docs surface, prior lifecycle blocker, stale-diff removal, residual NEO_MC_PRIMARY references, close-target metadata, targeted local tests, and GitHub CI. I found no new blocking concerns. Residual references in DeploymentCookbook.md and v13-path.md are deprecation/roadmap context, and ai/mcp/server/memory-core/config.mjs is an ignored local generated config, not part of the PR surface.

Test-Execution & Location Audit

• Changed surface class: code + docs + test deletion.
• Location check: pass; removed test was in the correct memory-core unit-test tree and no replacement file is required for the retired primary-flag behavior.
• Related verification run: npm run test-unit -- test/playwright/unit/ai/mcp/server/memory-core/Server.spec.mjs passed: 4/4.
• Additional verification: git diff --check origin/dev...HEAD passed.
• Findings: pass.

Contract Completeness Audit

• Findings: Pass. #10972 ACs map to the config, startup/disconnect path, and health/docs cleanup now visible in the diff.

CI / Security Checks Audit

• Ran gh pr checks 10978.
• Analyze, CodeQL, integration, and unit checks all pass.
• No pending or deep-red checks observed.

Metrics Delta

• [ARCH_ALIGNMENT]: 68 -> 92 — improved because the stale BaseServer lifecycle mismatch is gone; 8 points remain deducted only because this relies on surrounding daemon/lock substrate rather than proving end-to-end deployment behavior in this PR.
• [CONTENT_COMPLETENESS]: 70 -> 90 — improved because the active docs I flagged were reconciled; 10 points deducted for stale PR-body commit metadata not reflecting the final head, which is non-blocking because the author response provides the delta.
• [EXECUTION_QUALITY]: 58 -> 90 — improved because the branch is clean, targeted local verification passed, and CI is green; 10 points deducted because the deleted gate test naturally removes a direct queue-writing assertion and the deeper memory-core suite remains better trusted to CI in this sandbox.
• [PRODUCTIVITY]: 72 -> 95 — closes the intended #10972 retirement surface with only minor metadata/documentation residue.
• [IMPACT]: unchanged from prior review at 82 — still a meaningful cleanup of transitional Memory Core deployment state.
• [COMPLEXITY]: 66 -> 45 — lower after rebase because the PR now contains the focused retirement diff instead of stale M2 migration surface.
• [EFFORT_PROFILE]: unchanged from prior review at Maintenance — targeted retirement of an obsolete process-role flag.

Measurement Payload

• Static loaded surface: 56,799 bytes (pr-review skill/router, guide, follow-up template, CI audit, measurement methodology).
• Dynamic loaded surface: 28,869 bytes (git diff 18,775 + PR body/comments 8,208 + issue #10972 1,886).
• Total measured surface: 85,668 bytes.

Required Actions

No required actions — eligible for human merge.

Per §0: this is merge eligibility feedback only; merge execution remains human-only.