[DoodleBears]

Requested by Vain

Refs BotHarness/DeepSeekBot#1023 · Core PR · follows #31 (video)

Why the change

Let the external Consumer play a Lark voice message that a user quotes while mentioning the App, the same way #31 did for video.

Special things to note

• Merge risk: two-way door — revert the sourceVoiceAudio opt-in and the source-voice-audio-checked capability; small blast radius — only the opted-in parent-resource path in external-files.mjs; consumers that do not pass sourceVoiceAudio see no change.
• The Provider does not decode: it returns the parent's file_key as one audio/unknown attachment (voice.opus) plus voice: { transcript: 'unavailable', itemId, durationMs? }. Lark has no transcript, so none is claimed.
• Real E2E (2026-11-03, QA1 app, group "BotHarness QA1 Test"): a native audio message was quoted by a real Human mention; the Core Channel showed the voice with duration 4916 ms, the original read back byte-identical (156874cb…), and Chromium played the Host-decoded WAV with advancing time and seek.

Change outline

external-files.mjs   parentResources(): msg_type 'audio' when sourceVoiceAudio
                     → parentAudio(): exact file_key, optional duration, voice descriptor
                     readExternalFile(): re-check also accepts the audio parent
feishu-runtime.mjs   enrichExternal() forwards sourceVoiceAudio
multi-bot-controller consumeInbound({ sourceVoiceAudio }) + capability source-voice-audio-checked
docs/botharness-lark-audio.md   contract (EN/中文)

Tests: Feishu suite 791/791; new cases cover opt-in, mention requirement, exact-parent re-check, invalid duration, missing duration, other chat.

🤖 Generated with Claude Code