feat(web): pair a thread with an executor from the composer

[rynfar]

What

Adds the Pair control to the web composer, beside the model picker. One click opens it, one picks the executor's model, one turns the pair on.

• Off: the trigger reads "Pair". The panel holds the switch, a one-line explanation, and the executor model picker. The picker starts from the project's default delegation model, or opens on the first provider that actually has models.
• On: the trigger shows the executor's model, and a dot when the executor needs an approval or an answer. The panel shows what the executor is doing and links to its thread. The executor's model is fixed while the pair is on.
• The switch says why it cannot change: the lead is mid-turn ("Changes apply between turns."), no executor model is chosen, the lead is on Antigravity, or Pylon delegation is off. The last one matters because the server only gives a lead its pair tools while delegation is on, so a pair started without it would create an executor nothing could brief. Turning a pair off is never blocked by it.
• Turning on creates the executor thread under the deterministic pair id, in the lead's branch and worktree; if that id already exists as an archived executor it is unarchived instead. Turning off deletes an executor that was never briefed and archives one with history. An idle lead's session is stopped after a change so the next turn starts with the right tools and instructions.
• The control is not shown on a delegated thread.

How

The client only dispatches existing thread commands (create, unarchive, archive, delete, stopSession); nothing on the server or in the contracts changes. Decisions live in pure functions (pairControl.logic.ts, initialExecutorInstanceId), the hook (usePairControl.ts) wires them to commands, and PairControl.tsx renders. ChatView.tsx is untouched.

Verification

• vp test run on PairControl.test.tsx, pairControl.logic.test.ts, ProviderModelPicker.test.tsx: 36 passed.
• Web typecheck: 0 errors. Lint on the changed files: no new warnings. vp run knip:check: clean.
• In the browser against an isolated test server: the control renders beside the model picker; the panel opens; turning off deleted a never-briefed executor and the sidebar count dropped; the switch was blocked with "Choose an executor model first."; the picker opened on a provider with models; after choosing a model, turning on recreated the executor and the control read "Pair with Claude Sonnet 5: Waiting for a brief"; turning off again returned it to "Pair".
• Not run: the delegation-off reason in the browser (covered by unit tests only), and the unarchive path for an executor with history.

Known limits

• Abandoning a draft, or "Work locally" rotating a thread id, can leave an executor without a lead. Tracked in feat: pair mode, a linked lead/executor thread pair replacing fan-out delegation #622.
• Mobile has no Pair control yet.

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 4:48pm 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	13.9 KiB	−8 B (−0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.2 KiB	7.2 KiB	0 B (0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.7 KiB	6.7 KiB	−8 B (−0.1%)	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	−54 B (−0.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.2 KiB	7.2 KiB	−9 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.8 KiB	6.7 KiB	−45 B (−0.7%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	58.9 KiB	58.8 KiB	−88 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	10	8	−2 (−20.0%)	21	✅

Baseline: 2b6007f · PR result: d6dd251 · 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: 115.7 KiB
• Claude decoded thread snapshot: 116.3 KiB

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