fix(pair): no argument turns pair_await or delegated_thread_status into a poll

[rynfar]

What

Two guards so neither waiting tool can be used as a poll.

• pair_await: a lead may read a running executor instantly (maxSeconds: 0) twice in a row. A third instant read waits the whole cap instead. Any wait, or finding the executor not running, clears the count.
• delegated_thread_status: omitted or 0 reads the state; any positive waitSeconds waits the whole 45 seconds. It already returns the moment the child changes, so a shorter wait gains nothing.

Both tool descriptions say so.

Why

#640 made positive pair_await waits run to the cap after a live Codex lead looped on 10 to 20 second waits, but left two holes: a loop on maxSeconds: 0, and the same short-wait loop on the fan-out status tool. Each return from either tool is a model turn.

How

• pairAwaitPlan (pure) returns the wait budget and the next instant-read count; the handler keeps the count per executor in memory beside the protected-path record.
• delegatedStatusWaitSeconds (pure) and MAX_STATUS_WAIT_SECONDS move the cap into delegation/logic.ts.

Verification

• vp test run on apps/server/src/mcp/toolkits and apps/server/src/orchestration: 958 passed. Five tests were written first and seen failing: both pure rules, a handler test where the third instant read reports waitedSeconds: 45, and the fan-out test that used to return after 5 seconds.
• Typecheck t3: 0 errors. Format and vp run knip:check clean.

How it was built

This is the first change made through pair mode itself on a real install (nightly .201): a Claude lead paired with an Antigravity executor (gemini-3.8-flash-high) sharing one worktree. The lead wrote the contract and tests and listed the four test files as protectedPaths; one pair_handoff; Pylon woke the lead when the executor finished; pair_await reported protectedPaths: { checked: 4, changed: [] }; the lead read the diff, re-ran everything above, and wrote the docs and this text. One executor turn, no corrections needed.

Part of #622.

Model: Claude Fable 5.1 (lead) with an Antigravity executor, paired 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 10:08pm 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	−3 B (−0.0%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.2 KiB	7.2 KiB	+6 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.8 KiB	6.7 KiB	−9 B (−0.1%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.0 KiB	58.0 KiB	−44 B (−0.1%)	66.4 KiB	✅
Codex	Live turn messages	10	9	−1 (−10.0%)	21	✅
Claude	Total thread wire	14.0 KiB	13.9 KiB	−17 B (−0.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.2 KiB	7.2 KiB	−18 B (−0.2%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.7 KiB	+1 B (+0.0%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	58.9 KiB	58.9 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	9	9	0 (0.0%)	21	✅

Baseline: 4be539f · PR result: bbbc4c2 · 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.