feat(server): orchestrator v2 core fixes (upstream G1)

[rynfar]

Ports upstream group G1, orchestrator v2 core fixes, onto pylon in upstream order. Pylon adopted v2 from upstream de34391427. Pylon's own v2 changes are kept: the #989 test exceptions, the #990 Codex reopen fix, and the ThreadLaunchService delta.

Sources

Upstream	Outcome	How
fd7ee2c30a lint: forbid tests in for loops (#14921)	Adapted	Adds t3code/no-test-in-loop and rewrites every flagged loop to .each. Where Pylon's file differed from upstream, Pylon's content stays and only the flagged loops changed. Pylon-only suites the rule flags were rewritten the same way: Prime daemon, legacy v2 bridge, Codex absolute history, provider, MCP, desktop, client-runtime and the vcs/settings tests. Test counts and names are unchanged. Upstream's reconcileV2PreviewMigration.test.ts does not exist in Pylon and is dropped.
ca7df394ed steers keep the newest next-turn selection (#14725)	Adopted	Clean.
736130c2fd working timers no longer reset on background wakes (#15029)	Adapted	The Sidebar.logic.ts conflict keeps the Working section helpers (#991) and takes only the updated resolveWorkingStartedAt doc comment. The timer value reaches web and mobile through client-runtime.
ca4f84e702 threads settle when an agent merges their PR (#15024)	Adapted	Clean, with "T3" changed to "Pylon" in overview.md. The reactor only re-reads open PR snapshots. ThreadSettlementService still decides settlement, so sidebarAutoSettleOnMerge and the per-thread auto-settle opt-out still apply; its tests pass.
6108ef3d3d runs no longer get stuck (#15048)	Adopted, plus Pylon tests	Clean. New LegacyAdapterV2SessionManager.test.ts covers the Prime bridge going through the new session-manager release path, listed below.
ddcd310282 docs, dev scripts and CI catch up with V2 (#15041)	Partial	Adopted: the migrate-dev-db V2 rewrite (default source ~/.pylon-code/userdata/statev2.sqlite, never ~/.t3), removal of the Orchestrator interrupt TODO and TODO.md, the stale comments, the vite manual-runtime debt list for deleted V1 tests, the mobile-showcase seed, and the devices link. Adapted: glossary and providers docs, in Pylon wording. Skipped: migration comment renumbering (Pylon numbers V2 as 68/69), T3 AGENTS.md/CONTRIBUTING.md/test-t3-app fixture docs, shardWeights.json, the CI gate requiring transfer-report, and the composer-context doc hunk (Pylon's doc has no stale reference).

Pylon adaptations

• Prime bridge tests (apps/server/src/provider/legacy/LegacyAdapterV2SessionManager.test.ts). Three tests open the real bridge through ProviderSessionManager and wait on events and drain receipts, not sleeps:
  • Closing a session stops the legacy runtime and publishes the maintenance drain; the same test asserts a maintenance reservation fences new admission through the manager.
  • A release whose stopped record fails to commit still stops and drains, and the upstream retry records stopped once the store recovers.
  • A failed Prime daemon stream releases the session as error and drains.
• Local development reference: .agents/references/local-development.md now snapshots statev2.sqlite, noting that pre-v2 installs may only have state.sqlite (migrate-dev-db then fails safely with source-missing; pass --source).
• Codex reopen (fix(server): reopen Codex sessions whose app-server exited #990): OrchestratorReplayRecovery.integration.test.ts, the Codex adapter suite and the session-manager suite pass with the new release-record retries and ownership-read changes.
• Glossary: the orchestration table takes the V2 terms but keeps "Reactor" and "Runtime receipt", which Pylon's AGENTS.md still uses.
• migrate-dev-db: the doc and the test assertion now name statev2.sqlite.
• Typing: PrimeAgentDaemonEvents.test.ts imports an untyped Vitest runner, so its .each callbacks got explicit case types.

Verification

• vp test run src/orchestration-v2 src/provider/legacy scripts/migrate-dev-db.test.ts in apps/server: 1657 passed, 13 skipped, 1 environment failure. claude_result_is_error/claudeAgent fails only because this machine sets CLAUDE_CONFIG_DIR. It passes with env -u CLAUDE_CONFIG_DIR, as fix(server): reopen Codex sessions whose app-server exited #990 also noted.
• The new bridge suite passed three times in a row.
• Focused suites per source all pass: steering; ProjectionStore/runtimeLayer/threadExecution/Sidebar.logic; PullRequestSyncReactor/ProjectionSettlement/ThreadSettlementService; ProviderSessionManager/RunExecution/ProviderTurnStart/FoundationPersistence/Codex adapter/replay recovery; migrate-dev-db; mobile-showcase.
• Every test file the lint commit changed passes, across server, desktop, client-runtime, contracts, shared, web, relay, mobile and scripts. The lint rule's own test passes, 8/8.
• vp lint on every test file containing a loop reports no no-test-in-loop errors. vp fmt --check and vp lint on the changed files are clean apart from one existing warning in Orchestrator.ts.
• Typechecks report no error TS: vp run -F t3, @t3tools/web, @t3tools/desktop, @t3tools/client-runtime, @t3tools/contracts, @t3tools/shared, @t3tools/mobile, t3code-relay, effect-acp.

Not verified

• No UI capture. The only web change is a doc comment. The working-timer change is server and client-runtime data, checked by unit tests.
• migrate-dev-db was not run against a real database. Only its isolated tests ran.
• The full repository test suite was not run; CI covers it.

Part of upstream cycle #996.

🤖 Generated with Claude Code

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	4.9 KiB	4.9 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.7 KiB	3.7 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	20.4 KiB	20.4 KiB	0 B (0.0%)	29.3 KiB	✅
Codex	Live turn messages	2	2	0 (0.0%)	8	✅
Claude	Total thread wire	4.9 KiB	4.9 KiB	−41 B (−0.8%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.7 KiB	3.7 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	−41 B (−3.4%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	20.8 KiB	20.7 KiB	−41 B (−0.2%)	29.3 KiB	✅
Claude	Live turn messages	2	1	−1 (−50.0%)	8	✅

Baseline: 30ec43a · PR result: 660a328 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 106.1 KiB
• Claude decoded thread snapshot: 106.4 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.