fix(claude): between-turn status and resume handshake no longer touch the turn lifecycle

[rynfar]

Problem

Two Claude Code SDK messages that are not part of any turn were being projected as turn lifecycle, and both were traced from the nightly database and provider logs on 2026-09-04:

1. system/status between turns. When a background task finishes, the CLI re-invokes the model and reports status: requesting before any assistant output. The adapter mapped that to session.state.changed: running with no turn id, so the projected session sat in running with no active turn, sometimes for hours. The sidebar showed "Working" for an idle thread, and until fix(server): steer a turn only when one is actually running #264 the next message was admitted as a steer into nothing.
2. The resume handshake result inside a real turn. A resumed session replays system/init plus result(num_turns: 0). When that landed after the user's first prompt was already in flight, the adapter completed the user's turn after two seconds with no output. The actual answer then arrived as a synthetic turn. In the incident the visible turn showed nothing and the real work ran hidden.

Fix

• case "status" emits a lifecycle event only while a turn is open. Between turns it is the CLI's own request cycle, and the assistant output that follows opens its own synthetic turn.
• handleResultMessage ignores a zero-turn success result when the open turn has produced no items, no assistant text, and has no tool in flight. A real turn always makes at least one request, so that shape can only be the handshake. Failed or non-empty results complete as before. A log line (claude.turn.handshake-result-inside-turn) keeps the trigger measurable.

Upstream has identical code on both paths, so this is Pylon getting ahead rather than catching up. Upstream's #9135 (SDK 0.3.260, terminal-reason failure handling) is unrelated and still worth adopting through the upstream batch.

Verification

• Two new adapter tests, each watched failing first: the between-turn status produced a second lifecycle event, and the handshake result produced a second turn start and completion.
• vp test run on ClaudeAdapter.test.ts: 85 passed.
• vp run -F t3 typecheck: clean. vp fmt --check and vp lint on the touched files: clean.

Companion PRs: #264 (decider no longer steers unadmitted turns) and the follow-up that admits provider-initiated turns in ingestion.

Model: Claude Fable 5.1. Harness: Pylon (Claude Code provider).

---

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

[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	13.6 KiB	13.6 KiB	+10 B (+0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.1 KiB	7.1 KiB	+1 B (+0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.5 KiB	6.5 KiB	+9 B (+0.1%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	56.5 KiB	56.5 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	8	8	0 (0.0%)	21	✅
Claude	Total thread wire	13.6 KiB	13.6 KiB	−12 B (−0.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.1 KiB	7.1 KiB	−2 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.6 KiB	6.5 KiB	−10 B (−0.1%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	57.3 KiB	57.3 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	8	8	0 (0.0%)	21	✅

Baseline: bfc5631 · PR result: 6f4a03c · 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: 111.2 KiB
• Claude decoded thread snapshot: 111.9 KiB

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