Repository navigation
Resound v1.0.2 improves recording, recovery, and artifact safety after the hardening passes since v1.0.1.
Recording and recovery
- Keep recorder ownership until child processes close, reject overlapping starts, and share repeated stop results.
- Correct Discord speaker alignment and keep long-recording finalization responsive.
- Serialize shutdown and session operations; require every declared speaker track during recovery.
Data integrity and resource limits
- Reserve unique session directories and replace transcript and delivery files atomically per file.
- Prevent configured exports from replacing retained audio.
- Validate provider responses, nested manifests, transcript timestamps, and track timing before accepting output.
- Save consent and failure state for failed file imports or transcription.
- Bound subprocess and HTTP lifetimes, stream diagnostics, upload tracks sequentially, and convert PCM to WAV in bounded reads.
- Preserve large piped transcripts, honor supplied environment deadlines, and update the optional Strata command.
Documentation and dependencies
The README now gives direct setup, storage, and recovery guidance. Updated dependencies and the restricted Docker build context are included.
Compatibility and verification
Manifest schema 1.2.0, CLI flags, configuration keys, and provider interfaces remain compatible. Invalid or destructive inputs now fail explicitly. Per-file atomic writes are not a multi-file transaction or a power-loss guarantee.
Verified with 248 TypeScript tests, eight offline Python tests, 20 CLI checks, build/types/format checks, release metadata checks, and a dependency audit with zero advisories. CI covers Linux, macOS, the pinned Python receiver, and the container. Live Discord, devices, models, and remote providers still need the host's acceptance check.
See the changelog and self-hosting guide.
Full diff: v1.0.1...v1.0.2