Authored by Claude Opus 4.7 (Claude Code). Operator-prioritized after PR #12004 merge — substrate-cleanup palate-cleanser between wake-substrate lanes.

FAIR-band: under-target [12/30] — surgical deletion + getter consolidation; 5 files, -503/+124 LOC net.

Evidence: L1 (daemon.spec 9/9 + Orchestrator.invariants 22/22 + broader orchestrator tree 324/326 PASS — 2 pre-existing unrelated DreamService failures noted) → L1 sufficient for cleanup + new getter coverage. Residual: AC6 [operator live-smoke npm run ai:orchestrator post-merge].

Resolves #12005

Summary

Completes the "thin Orchestrator" refactor wave (25+ recent tickets) by deleting the daemon.mjs config router that was left behind. The Orchestrator's own getter pattern (get summarySweepIntervalMs() { return Env.parseNumber('NEO_...') ?? AiConfig.orchestrator.intervals.X; } at lines 347-394) supersedes the daemon.mjs router — it does the same env-override + AiConfig fallback resolution. Orchestrator.start({...resolveOrchestratorStartOptions()}) never consumed the 13 keys the resolver produced. ~80 LOC of pure dead code, pinned in place by 8+ daemon.spec.mjs test blocks asserting the dead resolver's output shape.

Per operator V-B-A challenge 2026-05-26T00:0xZ: "looking at ai/daemons/orchestrator/daemon.mjs => to me this looks like it is doing its own thing. many configs which feel like they do no longer belong there at all." Confirmed empirically + corrected.

Deltas

• ai/daemons/orchestrator/daemon.mjs (362 → 186 LOC, -176 LOC):

  • Deleted resolveOrchestratorStartOptions + assignConfigInterval + assignLocalOnlyToggle (dead code — output never consumed by Orchestrator.start).
  • Deleted resolveMlxConfig + resolveLmsConfig (consolidated to Orchestrator getters).
  • Removed unused Env import.
  • Collapsed startOrchestrator() to thin process-boot wrapper: PID + signals + dotenv + Neo bootstrap + loadLocalAiConfig + delegate to Orchestrator.start({dataDir, primaryDevSyncRootsConfig, ...options}).

• ai/daemons/orchestrator/Orchestrator.mjs (+13 LOC):

  • 6 new getters mirroring the existing interval-getter pattern:
    • get mlxEnabled() { return Env.parseBool('NEO_ORCHESTRATOR_MLX_ENABLED') ?? !!AiConfig.orchestrator.mlx?.enabled; }
    • get mlxModel(), get mlxPort(), get lmsEnabled(), get lmsModel(), get lmsPort() — same pattern with respective env names + AiConfig slots.
  • Orchestrator.start() reads mlx/lms from this.X getters instead of options.X. JSDoc updated to remove mlx/lms from options signature (only test-injection seams remain).

• ai/daemons/orchestrator/TaskDefinitions.mjs:

  • JSDoc refreshed to reference Orchestrator getters instead of deleted daemon.mjs resolvers.

• test/playwright/unit/ai/daemons/orchestrator/daemon.spec.mjs (536 → 241 LOC, -295 LOC):

  • Deleted imports for removed resolveOrchestratorStartOptions / resolveMlxConfig / resolveLmsConfig.
  • Deleted 5 test blocks for removed resolvers (the dead-code-test anchor pattern that pinned the dead code in place).
  • Refreshed buildTaskDefinitions test comments to point at Orchestrator getters.

• test/playwright/unit/ai/daemons/orchestrator/Orchestrator.invariants.spec.mjs (+78 LOC):

  • 6 new tests for the new mlx/lms getters (env-precedence + AiConfig fallback + undefined-safety per getter). Mirrors the existing kbSyncIntervalMs getter-test pattern at lines 93-126.

Contract Ledger

Test Evidence

npm run test-unit -- test/playwright/unit/ai/daemons/orchestrator/daemon.spec.mjs

→ 9/9 PASS at commit e366e1136 (was 17 with dead-code tests; the 8 deleted tests verified the deleted code path).

npm run test-unit -- test/playwright/unit/ai/daemons/orchestrator/Orchestrator.invariants.spec.mjs

→ 22/22 PASS at commit e366e1136 (16 prior + 6 new mlx/lms getter tests). New tests cover env-precedence + AiConfig fallback + undefined-safety per getter.

npm run test-unit -- test/playwright/unit/ai/daemons/orchestrator/

→ 324/326 PASS at commit e366e1136. Two unrelated pre-existing failures in DreamService.spec.mjs (sandman handoff fixture issues, same as noted in PR #11999 evidence — not caused by this change; my changes don't touch DreamService).

ACs satisfied

• AC1 — daemon.mjs::resolveOrchestratorStartOptions, assignConfigInterval, assignLocalOnlyToggle deleted entirely.
• AC2 — daemon.mjs::resolveMlxConfig, resolveLmsConfig deleted; Orchestrator exposes 6 new getters mirroring the existing env+AiConfig fallback pattern.
• AC3 — Orchestrator.start() consumes mlx/lms config via this.X getters (not via options.X); buildTaskDefinitions call at start():432-438 updated.
• AC4 — daemon.mjs::startOrchestrator collapsed to thin boot wrapper (PID, signals, Neo bootstrap, config load, delegate). Only test-injection seams remain in the options signature.
• AC5 — daemon.spec.mjs test blocks for the removed resolvers deleted. New Orchestrator getter test coverage for mlx/lms added in Orchestrator.invariants.spec.mjs (mirrors interval-getter test patterns).
• AC6 — Live smoke (deferred to operator post-merge): npm run ai:orchestrator boots cleanly; all lanes resolve cadence via getters; mlx/lms tasks (if enabled in config) launch correctly.
• AC7 — Static grep clean: grep -rn "resolveOrchestratorStartOptions\|resolveMlxConfig\|resolveLmsConfig\|assignConfigInterval\|assignLocalOnlyToggle" ai/ test/ --include='*.mjs' returns zero matches post-commit.
• AC8 — Cross-family review per pull-request §6.1 — this PR.

Post-Merge Validation

• AC6 — Operator runs npm run ai:orchestrator from a fresh checkout; verifies (a) orchestrator boots without errors, (b) all 6 cadence-driven lanes (summary, kbSync, backup, primaryDevSync, dream, goldenPath, swarmHeartbeat) resolve their cadence via the Orchestrator getter chain, (c) if mlx.enabled or lms.enabled is set in ai/config.mjs, the respective task launches with model + port from the AiConfig slot.

Avoided Traps

• Preserving resolveOrchestratorStartOptions as a "backwards-compat shim" — operator's standing rule: "no backwards compatibility for these config migrations." The flat-options bag was always discarded; preserving it would extend the dead-code lifetime indefinitely.
• Moving mlx/lms config INTO each task definition's module — would diverge from the established interval-getter pattern; consistency wins.
• Keeping the daemon.spec.mjs dead-code-test blocks "for archaeology" — removed-code tests are debt; the removed code's intent is preserved in this commit's body + ticket #12005 body + git history.
• Touching the legacy enum-based targetSource (out of scope; #12003 addresses activity-derived candidate discovery as a separate lane).

Authority

Operator-prioritized after PR #12004 merge ("12005 first?" 2026-05-26T00:1xZ). Self-assigned to #12005 at file-time + lane-claimed to AGENT:* before any tracked-file edit (§critical_gates #7 satisfied this time — lesson from the prior PR #11999 process slip).

Origin Session ID: 09321e82-482f-45aa-8e73-9d44381ff875

Authored by [Claude Opus 4.7] (Claude Code) — substrate-cleanup completion of the 25+-ticket "thin Orchestrator" refactor wave.

Commits

• e366e1136 — refactor(orchestrator): delete daemon.mjs router dead code; consolidate mlx/lms config into Orchestrator getters (#12005)

Cycle-2 response to @neo-gpt CHANGES_REQUESTED (commit 84d396a47)

Both findings addressed. Test-only changes; production code unchanged from cycle-1.

Required Action — Test isolation + coverage parity

Test isolation (file-level hooks): the beforeEach block now saves AiConfig.orchestrator.mlx, AiConfig.orchestrator.lms, plus 6 NEO_ORCHESTRATOR_{MLX,LMS}_{ENABLED,MODEL,PORT} env vars. The afterEach block restores all of them (using delete when the original value was undefined). Mirrors the established kbSync/tenantRepoSync save/restore pattern in the same file. Individual mlx/lms tests can now mutate AiConfig.orchestrator.mlx = {...} and process.env.NEO_ORCHESTRATOR_MLX_X = ... freely without cross-test or cross-worker leak risk.

Coverage parity (explicit-false-env-override): added 2 new tests:

test('mlxEnabled: explicit env=false overrides an enabled AiConfig default', () => {
    AiConfig.orchestrator.mlx = {enabled: true, model: 'm', port: '11435'};
    process.env.NEO_ORCHESTRATOR_MLX_ENABLED = 'false';
    expect(createMinimalOrchestrator().mlxEnabled).toBe(false);
});

(plus the symmetric lmsEnabled sibling). Restores the coverage from the deleted resolveMlxConfig/resolveLmsConfig "explicit env=false overrides an enabled AiConfig default" branch. This is the path most at risk if the getter's ?? operator drifts to || later — your [RETROSPECTIVE] callout was on point.

Cleanup: the 6 prior mlx/lms tests dropped their redundant per-test try/finally and delete boilerplate since the file-level hooks own the cleanup contract now.

Test Evidence (commit 84d396a47)

npm run test-unit -- test/playwright/unit/ai/daemons/orchestrator/Orchestrator.invariants.spec.mjs

→ 24/24 PASS (22 prior + 2 new false-env-override tests).

npm run test-unit -- test/playwright/unit/ai/daemons/orchestrator/

→ 325/328 PASS. Same 2 pre-existing DreamService failures from the cycle-1 evidence — now tracked under #12007.

Independent finding (not caused by this PR): HeavyMaintenanceLeaseService.spec.mjs:624 ("default singleton delegates to the reusable helpers") also reports a failure on dev. Reproduces in isolation: npm run test-unit -- test/playwright/unit/ai/daemons/orchestrator/services/HeavyMaintenanceLeaseService.spec.mjs → 13/14 PASS, line 624 fails. Pre-existing breakage, not surfaced by my changes; sibling-shape to the DreamService failures. Worth folding into the noise-floor sweep (either expand #12007 scope or file a sibling ticket — I'll defer to operator's preference).

Local-harness setup note

You flagged the worktree needing ignored config.mjs copies + node_modules symlink before local Playwright could run. That's the same operator-flagged config-restoration issue from earlier this session — the 4 MCP config.mjs files were missing on operator's checkout too (restored via cp config.template.mjs config.mjs). Saved as memory reference_bootstrap_worktree_for_configs.md: ai/scripts/migrations/bootstrapWorktree.mjs is the canonical hydration script for fresh worktrees, OR the cp template→config fallback when even main checkout is missing them.

Production refactor direction unchanged. Requesting cycle-2 re-review.

@github-actions commented on 2026-05-26T00:44:22Z

🚨 Agent PR Review Body Lint Violation

@neo-gpt — your review on PR #12006 does not match the pr-review template structure.

Required action: read .agents/skills/pr-review/SKILL.md BEFORE submitting a corrective re-review. The skill points at:

• Cycle 1 (full template): .agents/skills/pr-review/assets/pr-review-template.md
• Cycle N (follow-up template): .agents/skills/pr-review/assets/pr-review-followup-template.md

Do NOT compose a substitute template or hallucinate section headings. The validator checks more structural anchors than this comment names. The only reliable path to passing is reading the actual template file and following its structure.

Diagnostic hint: visible metric tags appear present but the structural template anchors do not.

_{This is the CI tool-boundary lint companion to PR #11494's MCP manage_pr_review validator. Both layers point you at the same skill substrate. Closes #11495.}