fix(prime): reconcile delivered user messages in native turns

[rynfar]

Native Prime could finish a request internally while Pylon failed the first turn with “transcript continuity could not be verified.” A complete replay snapshot can contain the delivered user message before its incremental event reaches the adapter.

Reconcile that single boundary only when the exact submitted text and ordered image signatures match, the current turn has not advanced or queued steering, and the existing complete replay, lifecycle, generation, and proof checks pass. Unknown replay and adopted turns without submission evidence remain rejected.

Validation: 408 focused native tests passed, followed by the final 22-case regression matrix; server typecheck and targeted lint passed. A real native browser turn completed in 13 seconds with both tool calls, and Supervised denial left the requested file absent. One independent adversarial checkpoint review passed. Remaining integrated approval, cancellation, restart, and mobile checks belong to the ongoing Prime first-class plan. Rebase onto current pylon preserved all three patches exactly (git range-diff).

Implemented and reviewed with GPT-6 in the Codex harness in Pylon.

---

^{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 11, 2026 2:12am UTC

[rynfar]

The native continuity fix now completes the original directory/README request in 13 seconds with two tool calls. Supervised denial leaves the file absent; approval in a fresh thread creates the file and exposes its checkpoint diff.

The independent N1 checkpoint review passed. The broader native baseline has a separate follow-up/Stop stall under investigation; these captures do not claim that flow passes.

native-first-turn-after.webm

[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	14.1 KiB	14.0 KiB	−35 B (−0.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.2 KiB	7.2 KiB	−5 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.9 KiB	6.8 KiB	−30 B (−0.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.8 KiB	58.8 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	14.0 KiB	−42 B (−0.3%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.2 KiB	7.2 KiB	+5 B (+0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.8 KiB	6.8 KiB	−47 B (−0.7%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	59.7 KiB	59.6 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	9	8	−1 (−11.1%)	21	✅

Baseline: ac8d87b · PR result: 81dfcbb · 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.