[AksharP5]

Disk errors while assembling transcription audio currently produce uncaught stream errors even when the caller catches the helper's promise. That can escape the workflow's intended fallback to post-mux transcription.

The helper now uses Node's pipeline() to consume the existing ordered download batches and handle output errors and completion. Two regression cases cover a pending-download output-open failure and a real Linux ENOSPC write. They run in the existing Recording Reliability workflow.

Closes #2938

Same Linux/Node scenario	Before	After
Await and catch a write to /dev/full	Caught ENOSPC, then an uncaught stream error; exit 1	Caught ENOSPC; exit 0
Output parent missing while fetch is pending	Uncaught ENOENT event	Promise rejects with ENOENT
Five focused Node checks	Suite fails	5 pass

Verified against current main 1965b3179 on Node 26.11.1. The Node checks also cover segment ordering, the concurrency bound, empty input, and HTTP failure details. Scoped Biome, strict helper typecheck with Node 20 declarations, workflow YAML parsing, and whitespace checks pass.

The two repository regression tests also pass under Bun with a temporary Vitest import adapter. Full Vitest and the complete hosted workflow were not run locally because this isolated checkout has no installed application dependencies. No deployment crash is claimed. Fetch batching, the byte limit, returned byte count, and the buffer-based helper are unchanged.

Confidence Score: 5/5

The PR appears safe to merge; no actionable issues were found.

Summary

Replaces manual stream writes with pipeline() so disk errors reject the promise and reach the existing transcription fallback.

• Segment audio assembly now reports output errors through its promise.

_{Reviews (1) · Last reviewed commit: "fix(web): handle disk errors during tran..." · Reviewed by Greptile}

[vercel]

@AksharP5 is attempting to deploy a commit to the Cap Software Inc Team on Vercel.

A member of the Team first needs to authorize it.