Specify a voice-note command without inventing media support

An engineering guide to capture consent, media lifecycle, accessibility, idempotency, and capability negotiation for an unevidenced LinkedIn message type.

Educational research only. LinkedIn account access remains decision-gated. No DewEngine LinkedIn connector is approved or implemented.

Talk to us ↗Read technical guides →

Current boundary

Voice notes are not declared as a DewEngine capability

The LinkedIn target names messages and message.send, but it does not list an attachment resource, audio upload method, codec contract, duration limit, or voice-message event. A generic message target cannot prove a media subtype. The entire LinkedIn path is decision-gated, so this guide defines the product questions rather than offering a working payload or provider instructions.

Return unsupported until a media contract is evidenced
Do not infer voice support from text-message naming

Capture

Make microphone intent and destination explicit

An application should request microphone permission only when the user starts recording and should show the acting account and intended conversation throughout capture. Recordings need stop, preview, delete, and re-record controls. If the conversation or account changes before approval, invalidate the draft so private audio cannot be delivered to a newly selected recipient by accident.

Bind the recording to account and conversation IDs
Delete abandoned temporary media on a short, tested schedule

Media pipeline

Separate recording, storage, upload, and message creation

A robust design validates size, duration, format, and decodability before a provider attempt. Temporary media belongs in encrypted object storage with checksums, content type, ownership, retention, and antivirus or safety policy where appropriate. Upload acknowledgement and message acknowledgement are different states. If either contract is absent, the command should fail clearly rather than pretending a text send covers it.

Use opaque object references instead of embedding audio in logs
Garbage-collect media after definite completion or expiry

Accessibility

Provide a usable alternative to audio

Recipients and senders may be unable or unwilling to use voice. Let the sender review a transcript if one is generated, but label it as derived and allow correction. Do not send the transcript automatically as a second message. Playback needs accessible controls and duration. A text alternative should remain a user choice, not a silent fallback that changes the intended communication.

Keep audio and transcript permissions aligned
Avoid retaining transcripts longer than the underlying purpose

Dispatch truth

Fence duplicates across upload and send ambiguity

Persist a command and media hash before external work. A lost upload response or message response can leave an indeterminate outcome. Automatic re-upload or resend may create duplicate artifacts or messages. Reconcile only through an approved provider read contract and use stable attempt records. Account restriction, authorization loss, or missing local media should stop immediately with a truthful state.

Do not label upload success as message delivery
Prevent a retry from changing the approved recording

Release checklist

Require media-specific conformance

Legal and provider authorization must precede implementation. Evidence would then need eligible-account support, exact formats and limits, empty and corrupt media, interrupted upload, duplicate attempt, revoked session, restriction, cross-tenant object substitution, playback, transcript privacy, deletion, retention, and kill-switch tests. Until a manifest includes proven media operations, DewEngine should not advertise LinkedIn voice-note sending.

Test content deletion from objects, caches, and backups
Publish unsupported clients or account products precisely

Questions

Before you build.

Can message.send be assumed to accept LinkedIn audio?+

No. A generic method name does not establish media upload, supported formats, size limits, or recipient behavior. Voice notes need a separate, evidence-backed capability contract.

Should a generated transcript be sent with every voice note?+

Not automatically. It is derived personal content and may contain errors. Let the sender review and choose whether to include a corrected text alternative, with separate retention and privacy controls.

Build with us

Does this match the workflow your users need?

DewEngine is in development. Real use cases decide what ships first.

Share your use case ↗