feat(delegation): skip the wake for a completion the parent already read

[rynfar]

What

A parent that waits for a child inside its own turn and gets the finished result was still woken by the follow-through reactor once that turn ended. The wake carried nothing new and cost a full parent turn each time. This removes it.

Seen live with a paired Claude lead: it called pair_await, verified the executor's work, answered, and was then woken to say "already completed".

How

The reactor already skips a child whose stored delegation.child-state receipt has the current noticeKey and baseline: true. The tools now write that same receipt.

• consumedDelegationObservation (pure): an observation counts as read only for a finished attempt: completed, error, or interrupted. A running child and a child that needs the user return null, so those wakes still happen.
• delegationObservationReceipt (pure): the one place the receipt's id and payload are built. The reactor uses it too, so the two writers cannot drift.
• markDelegationObservationConsumed: dispatches one thread.activity.append on the parent with a deterministic command id. It never fails a tool call; if the write fails, the cost is the old redundant wake.
• Called from pair_await once the executor is not running, and from delegated_thread_status and delegated_thread_result.

Activities are stored by id, so the tool's receipt replaces what the reactor recorded while the parent was busy, and either order of the two writes ends the same way. No contract, decider, projector, or migration changes.

Verification

• vp test run on delegationFollowThrough.logic.test.ts, DelegationFollowThroughReactor.test.ts, and the pair and delegation toolkits: 172 passed. New cases: the pure rules, one receipt per finished pair_await / delegated_thread_status and none for a running child, and a reactor test where a child finishes while the parent is busy, the receipt is rewritten with baseline: true, and the idle parent is not woken.
• Typecheck t3: 0 errors. vp run knip:check: clean.
• Live, isolated server, Claude lead with a Claude executor: before, one brief produced two assistant turns and a delivered wake; after, one brief produced one assistant turn, the receipt is stored with baseline: true, and no wake was delivered.

Built test-first: the contract and the failing logic tests were written by the lead and left unchanged (blob 6131d97c), the implementation and handler tests by an Antigravity executor, the reactor test, review, live check and prose by the lead.

Part of #622.

Model: Claude Fable 5.1 (lead) with an Antigravity executor, in Pylon via the Claude Code harness.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
pylon-marketing	Ready	Preview	Sep 18, 2026 5:26pm UTC

[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	14.0 KiB	14.0 KiB	−8 B (−0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.2 KiB	7.2 KiB	+4 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.7 KiB	6.7 KiB	−12 B (−0.2%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.0 KiB	58.0 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	9	9	0 (0.0%)	21	✅
Claude	Total thread wire	14.0 KiB	13.9 KiB	−46 B (−0.3%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.2 KiB	7.2 KiB	+6 B (+0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.7 KiB	−52 B (−0.8%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	58.9 KiB	58.8 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	9	8	−1 (−11.1%)	21	✅

Baseline: 6468f56 · PR result: d326bf3 · 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: 115.6 KiB
• Claude decoded thread snapshot: 116.3 KiB

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