fix(acp): keep one answer through tool progress

[rynfar]

ACP tool progress updates currently complete the active assistant item, which splits one streaming answer into separate items. This is visible when a background command reports progress while the agent is writing a markdown table.

Track recently shown tool-call IDs and close the assistant segment only for a newly shown tool. An active call remains recognized if its ID ages out of the bounded recent-ID cache; terminal and repeated updates keep their emitted tool state without splitting prose. The change selectively ports T3 source e759847f932c0de08261ec98a5fac3bae037120f (#13386) into Pylon's ACP runtime, without changing wire contracts.

Pylon's Prime adapter also used every tool update as a final-response boundary. It now consumes the ACP runtime's explicit boundary decision, so a background tool completing after the final text cannot cause a false missing-final-response warning. A new tool after text still resets that response state.

Focused validation on 3d4fa44c10e2957f3a7c086ef5e8633b123e2ab1: vp test run apps/server/src/provider/acp/AcpJsonRpcConnection.test.ts apps/server/src/provider/Layers/PrimeAgentAdapter.test.ts (27 passed), vp run --filter t3 typecheck, scoped vp lint, and scoped vp fmt --check. The regressions cover progress, completion, repeated completion, an active tool aged out after 256 later calls, and Prime settlement with no text after the background completion. The existing Prime tool-only test still checks the missing-final-response warning.

Tracking: #865. Independent adversarial review and final-head CI remain pending.

Model: GPT-6 (Codex in Pylon harness).

[vercel]

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

1 Skipped Deployment

Project	Deployment	Actions	Updated
pylon-marketing	Ignored	Preview	Sep 27, 2026 5:27am 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	−28 B (−0.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.2 KiB	7.3 KiB	+12 B (+0.2%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.8 KiB	6.7 KiB	−40 B (−0.6%)	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	14.0 KiB	−9 B (−0.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.2 KiB	7.3 KiB	+9 B (+0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.8 KiB	6.7 KiB	−18 B (−0.3%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	58.9 KiB	58.9 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	10	9	−1 (−10.0%)	21	✅

Baseline: 415c668 · PR result: 3d4fa44 · 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: 116.4 KiB

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