[lukemaj]

What: Keep the same-task follow-up delivery adaptation; upstream fixes the notification bug with a different task lifecycle.
Why: Removing the carried non-test behavior breaks 13 behavioral tests plus the continuation-guidance test on the already-absorbed upstream base.
So what: This registry-only PR clarifies the retained purpose and watched files; the #411 absorption must keep this adaptation unless it deliberately migrates the fork's task lifecycle.

Retire check, 2026-10-18

This supports the Objective's delegated Drafters by preserving follow-up delivery without changing their task identity.

• Starting branch/workspace: chore/212-retire-child-follow-up-wake, /home/luke/.t3/worktrees/chromeria/chore-212-retire-child-follow-up-wake; clean starting HEAD dfb9235811e8d9b76f4713c383301e1a220364ce, matching live origin/main.
• Read Delegated child follow-up turns wake the parent #212, the full fix(server): a delegated child's finished follow-up turn wakes its parent #213 diff, registry rules/entry, fork maintenance guidance, and chore(fork): daily upstream absorption through d0918f1659 #411 comment 6249777393.
• Actual fixing source: pingdotgg/t3code#17696, merged 2026-10-10T05:10:11Z as c572cb71d2711d8cdb4ff7e0ba88af5e9a23aa81. Read the complete 14-file diff. #13490 closed at 05:10:12Z. Candidates #14030, #15004 and #13338 are unmerged.
• git log upstream/main --oneline --grep='17696' finds the fix. Both git merge-base --is-ancestor c572cb71d2711d8cdb4ff7e0ba88af5e9a23aa81 f798cb299ac9eab9d51ef63b5159de14a2b6aa18 and the same command against d0918f1659f4c19172796c8d1910ca6e4bc18b51 exit 0. No detached worktree or write in the chore(fork): daily upstream absorption through d0918f1659 #411 worktree was needed.

Equivalence boundary

Upstream's recordSubagentFollowup creates a fresh delegated task, node and turn item after the original result. Each task has its own pending delivery. It excludes notifications, schedules and restart-continuation intake. Its completed-follow-up regression also finishes a Monitor run and verifies no extra result transfer or task mutation.

The fork deliberately changed this during absorption, recorded in 04770c9147e8940641edb7a818f93803b371564a (wip: preserve native follow-up identity and fork result delivery). Current recordSubagentFollowup binds later requested runs to original.id, and childThreads/delegatedFollowups.ts selects and recovers results per run. The retained #212 contract keeps the first summary/result stable while showing later results as latestTerminal*. With that same-task lifecycle, upstream's normal completion planner sees an acknowledged/delivered result and will not reserve another wake. The carried rearming and stable-result handling are still required.

Removal proof

Used Node 24.15.0 and pnpm 11.10.0 through a disposable npm execution environment because this shell had Node 26 and no pnpm. Installed dependencies only in this task worktree. No versions changed.

Exact focused command, run before removal, with removal, and after restoration:

npm exec --yes --package=node@24.15.0 --package=pnpm@11.10.0 -- pnpm exec vp test run \
  apps/server/src/orchestration-v2/DelegatedCompletionDelivery.test.ts \
  apps/server/src/mcp/delegatedTaskContinuation.test.ts \
  apps/server/src/mcp/OrchestratorMcpToolkit.integration.test.ts \
  packages/provider-core/src/server/orchestrationInstructions.test.ts

• Baseline: 4 files, 35 tests passed.
• Removal: 3 files failed, 1 passed; 14 tests failed, 21 passed. This includes 13 behavioral failures and 1 guidance assertion.
• The original once-per-follow-up regression reaches the second transfer but fails expected 'acknowledged' to equal 'claimed'. Previously stopped and wait-mode parent regressions fail the same way. MCP integration observes the original summary overwritten with the follow-up response. These are behavior failures, not compilation or dependency failures.
• Removal touched only the carried non-test logic in Orchestrator.ts and the three instruction sites, following the instruction-file moves since fix(server): a delegated child's finished follow-up turn wakes its parent #213. Reverted stable-result handling, delivery rearming, forced async policy, parent-cohort reopening, and follow-up suppression of original node/card rewrites. Retained upstream per-run deduplication and unrelated later restart/Stop/Prism hooks. All tests remained byte-for-byte unchanged. This is a removal experiment inside the current fork, not a claim to have run a pristine upstream checkout or migrated the fork back to fresh tasks.
• Restored all four non-test files byte-for-byte; git diff --quiet exited 0 before the registry edit. The delivered diff changes only the registry purpose, shared-file coverage and keywords.
• Monitor exclusion and duplicate replay are exercised by the passing baseline and restored original regression. The failed removal run stops before that test's Monitor assertions, so it is not claimed as independent negative-wake proof.
• Restored run: 4 files, 35 tests passed, including duplicate replay and Monitor exclusion.
• Registry check: node scripts/fork-features.mjs check passes, 42 features. git diff --check passes.

No runtime change, server, browser, live database, Rocky work or repo-wide check. No version bump. No feature-map part is added, removed or shipped; the registry entry remains carried. No website impact: maintainer registry metadata only.

Elon record

Requirements and who asked: User requested an evidence-based retire check of #213, keeping its regressions and Monitor exclusion; #212 also requires stable first-result semantics.
Deleted: The proposed wholesale retirement was rejected by removal proof; no new task model, runtime patch or test rewrite is needed.
Bottleneck: Same-task completion-delivery state, not missing upstream code: upstream's fresh-task design is already in both bases.
Checked myself: Read upstream #17696's entire diff, verified both ancestry relationships, inspected the fork's identity-preserving intake, and ran the retained tests before removal, with the edit removed, and after restoration.

Refs #212

Model and harness: GPT-6.1-Sol via Codex in T3 Code.

[lukemaj]

Independent review, relayed unchanged from the delegated reviewer (GPT-6.1-Sol via Codex):

PASS. Reviewed head 3a8e3697f14330f8a2eb17f871a19a83a8859ea3 against base dfb9235811e8d9b76f4713c383301e1a220364ce; no actionable findings.

• Only docs/fork-features.md changes. Runtime and tests remain byte-identical to base.
• Retention is justified: upstream creates fresh tasks; the fork reuses the original task. Stable first-result behavior is an explicit Delegated child follow-up turns wake the parent #212 acceptance criterion.
• Added shared paths accurately cover intake, result selection and recovery while preserving existing primary ownership.
• Inspected proof supports 35 passing baseline/restored tests and 14 removal failures: 13 behavioral, one guidance. Monitor exclusion and duplicate replay are covered by the passing regression; removal fails before its Monitor assertions.

Evidence limits: reviewed supplied logs and patch without rerunning tests. No pristine-upstream or live verification claimed. No edits or GitHub posts made.

[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	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 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	20.9 KiB	20.9 KiB	0 B (0.0%)	29.3 KiB	✅
Codex	Live turn messages	2	2	0 (0.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	−7 B (−0.1%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	−7 B (−0.6%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.2 KiB	21.2 KiB	0 B (0.0%)	29.3 KiB	✅
Claude	Live turn messages	2	2	0 (0.0%)	8	✅

Baseline: dfb9235 · PR result: 3a8e369 · 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: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

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

[lukemaj]

Review round 1 at exact head 3a8e3697f14330f8a2eb17f871a19a83a8859ea3: both required independent reviews APPROVE with no findings.

• Grok 4.7 (medium): same-task claim verified against upstream c572cb71 (fresh task per follow-up) and the fork's original.id binding and planner gate; registry check OK (42 features); no smaller safe deletion.
• Codex gpt-6.1-sol (high): same conclusions; did not rerun tests, so the 14-failure removal result remains author evidence.

Ready for the final train review and the user's merge decision.

🤖 Claude Opus 5.5 planner via Claude Code in T3 Code