Skip to content

Latest commit

 

History

History
45 lines (30 loc) · 3.88 KB

report.md

File metadata and controls

45 lines (30 loc) · 3.88 KB

PR4196 live recording verification

Source ad71d0391615ee336f0157941f473e204cfd0be8. Verified 2026-11-03 after rebasing onto current main. The repository remained clean and unchanged.

Own pnpm install --frozen-lockfile, pnpm build, pnpm build:android, and worktree-scoped pnpm clean:daemon passed. The test used SDK 0.68.1 and quality 5 for both the daemon and independent recorder.

Commands: node verify.mjs android, then node verify.mjs ios. The attached script reads the operator's existing Limrun credentials, creates its own instance with a 45-minute hard timeout, and always closes its session, daemon, forwarding proxy and instance. It edits no repository source.

Platform Runtime Instance CLI result
android Android 15/API 35, limdroid_x86_64_only android_euna_01m6h7s87mf878054h753zm8w0 13 passed, 1 expected injected 503 failure
ios iOS 26.4, iPhone 16 Pro ios_euna_01m6h7t5h8f87b4z7zyrcw5c85 13 passed, 1 expected injected 503 failure

Command flow

Each flow runs connect limrun --platform <platform>, opens Settings, and takes a screenshot. An independent SDK recorder starts with an explicit ID. The daemon then records/stops normally, records/stops with one download failure followed by retry, and starts a third recording finalized by close. Each invocation's full arguments, JSON output, stderr, exit code and timestamps are in its transcript.

Retry and isolation proof

A localhost HTTP/WebSocket proxy forwards the daemon's unchanged requests to the real Limrun instance. It records recording control messages and MP4 downloads. For exactly one download, it responds 503 before forwarding. The first record stop reports that failure. The second succeeds using the identical URL. The captured control stream has exactly one remote stop for that recording across both commands.

After start, normal stop, failed download, retry and close, an incompatible-quality probe is rejected by the shared encoder. Android additionally returns 403 when the still-active external recording file is requested. After close, a separate visual change succeeds, and stopping the original external recording ID still succeeds. The downloaded external MP4 spans the daemon operations. The probes are behavioral observations, not fixed-delay assumptions.

Outputs and cleanup

Platform File Codec/dimensions Duration Bytes
android normal.mp4 h264 360×808 1.608222s 27003
android retry.mp4 h264 360×808 1.382322s 28114
android close.mp4 h264 360×808 1.621644s 45177
android external.mp4 h264 360×808 10.003322s 283172
ios normal.mp4 h264 602×1310 0.871667s 174695
ios retry.mp4 h264 602×1310 0.758333s 51092
ios close.mp4 h264 602×1310 0.901667s 53925
ios external.mp4 h264 602×1310 7.638333s 753478

All eight MP4s passed ffprobe validation and a full ffmpeg -v error -i <file> -f null - decode. File hashes and exact commands are in the summaries.

  • android: scoped daemon stopped; public API confirmed instance termination at 2026-11-03T17:37:33.810Z.
  • ios: scoped daemon stopped; public API confirmed instance termination at 2026-11-03T17:38:02.028Z.

The one initial harness setup failure was a missing root-level ws resolution, before any instance was created. The script now resolves ws from the SDK's own dependencies. No runtime/product failure occurred.

Publication must exclude each instance-private.json and the state/ directories. Published transcripts redact API and instance tokens, including URL query tokens.

The recordings are included for inspection: Android normal, retry, close, external; iOS normal, retry, close, external.