[Narrator]

Domscribe now has a repeatable Docker Cloud Sandboxes smoke runner before we introduce an autonomous issue-fixing agent. The TypeScript runner uses @docker/sandboxes to launch a shell kit, build a pinned public commit in an isolated worktree, run the existing Chromium replay harness, and retain bounded local evidence. Success requires the eight browser assertions and confirmed deletion of the same sandbox UID; EOF, exit 0, missing evidence, cancellation, and cleanup failure cannot silently pass.

The runner uses SDK-managed OAuth or environment PAT auth without uploading host workspaces or credentials. Execution, transfer and cleanup each have deadlines, with automatic sandbox deletion after 30 minutes as backup. CI runs offline lifecycle regressions, types, lint and shell syntax only; it does not allocate cloud resources. The private SDK dependency is pinned to 0.53.0 with a scoped undici 6.29.1 security override. Usage and limits are documented in docs/docker-cloud-replay.md.

Validation on October 25:

• Real SDK/service run of source 92e939a: React 18.3.1 / npm 11.19.1 / Vite 7.3.7, Node 24.22.1, Chromium 156.0.8078.4. All eight browser checks passed: exact source/props/state, stale-handle refusal, fresh-session reordered-row annotation replay, ambiguous stop, intended-customer-only update, and production instrumentation removal. Zero application console/page/network errors. Evidence downloaded; deletion confirmed. About two minutes from allocation through cleanup.
• A second real SDK probe deliberately cancelled after receiving streamed process output. It recorded an interrupted run with no exit event and confirmed deletion of the same sandbox UID. Both accepted cloud resources were deleted.
• 17 offline cloud-runner tests and two existing process cleanup regressions pass, including cancellation during streaming, UID replacement, uncertain creation, failed/missing evidence and cleanup failures. Native launch initially rejected an empty stored-secrets list; the optional feature is now omitted, with a regression test.
• Uncached configured lint/test/build/typecheck targets passed for 16 projects, including 2,114 source tests. Standalone SDK types, ESLint with zero warnings, shell syntax, formatting, frozen lockfile install and audit passed; audit reports no known vulnerabilities. Native cloud setup built all artifacts uncached from its frozen lockfile.
• Independent agent review found no material issues; suggested mid-stream cancellation coverage was added and re-reviewed. Required GitHub review is separate.

Scope: one fresh controlled cloud consumer using locally packed source artifacts, rather than the published 0.6.1 APIs or a full cloud framework matrix. This does not implement GitHub issue ingestion, model-agent execution, fix PR creation, publishing or deployment. A realistic agent bug-fixing experiment is next.

[nx-cloud]

View your CI Pipeline Execution ↗ for commit 83873fe

Command	Status	Duration	Result
nx run domscribe-test-fixtures:install-fixture-...	✅ Succeeded	12s	View ↗
nx run domscribe-test-fixtures:integration--web...	✅ Succeeded	2m 52s	View ↗
nx run domscribe-test-fixtures:integration--web...	✅ Succeeded	2m 48s	View ↗
nx run domscribe-test-fixtures:integration--web...	✅ Succeeded	2m 5s	View ↗
nx run domscribe-test-fixtures:integration--web...	✅ Succeeded	2m 5s	View ↗
nx run domscribe-test-fixtures:install-fixture-...	✅ Succeeded	41s	View ↗
nx run domscribe-test-fixtures:install-fixture-...	✅ Succeeded	38s	View ↗
nx run domscribe-test-fixtures:integration--vit...	✅ Succeeded	43s	View ↗
Additional runs (18)	✅ Succeeded	...	View ↗

💡 Verify your cache is correct by running tasks in a sandbox. Read docs ↗

---

☁️ Nx Cloud last updated this comment at 2026-10-25 23:58:30 UTC