release: cut v1.3.55

[dofastted]

Summary

• classify client cancellation as lifecycle completion without account/session cleanup
• keep worker streams open through kin_job_done / trailers after message_stop
• bind parent/child sessions to one family VM with independent seats and generation recovery
• preserve parent/root/device metadata through Claude/newapi paths
• ship verified Rust kernel and cli-node binary assets

Verification

• node --test test/unit/failover-runner.test.mjs test/unit/pool-scheduler.test.mjs test/unit/sticky-router.test.mjs test/unit/go-worker-client.test.mjs test/unit/error-class.test.mjs
• 170 focused Node tests passed
• vm-13 exact cli-node/kernel: normal Haiku/Sonnet pong, parent+child two-seat run, child cancellation, parent message_stop, ready_slots=20, wedged_slots=0

Deployment gate

• Merge and publish v1.3.55 release assets before HostDzire deployment.
• Existing unrelated worktree changes are excluded: src/lib/vm/wrap-cli-runtime.mjs, web/pnpm-workspace.yaml.

---

Note

Medium Risk
Changes how usage is attributed on the OpenAI Responses proxy path, which can affect logged costs and cache token stats; UI-only changes elsewhere.

Overview
Improves OpenAI/Codex billing visibility and usage capture on the Responses upstream path.

Backend: While proxying OpenAI Responses SSE, the pool now parses usage from stream lines via usageFromSseLine and attaches it to inference results (including when the assembled body omits usage). Request logging gains a unit test that input_tokens_details.cached_tokens / cache_write_tokens map to cache read and creation counts.

Panel (Codex VMs): VM overview adds a 计费窗口 card (5h / 7d / 今日) with official cost, request counts, in/out tokens, and cache read/write. List and card views show richer 今日 and 5h token breakdowns, embed compact quota query / reset-credit actions on Codex rows (always visible in the list actions column), and tweak compact button labels in OpenaiQuotaActions.

^{Reviewed by Cursor Bugbot for commit 79d5630. Configure here.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_5cad2512-da6e-4139-b7e2-4a846f0977ea)