fix(feishu): map audio messages to VOICE for STT transcription (#30822)

[Linux2010]

What broke

Feishu voice messages (Opus/OGG) are classified as MessageType.AUDIO. After the Telegram audio/voice split, MessageType.AUDIO is treated as a file attachment and skips STT transcription entirely. Since Feishu has no separate VOICE message type, all voice messages end up being silently dropped from the STT pipeline, breaking auto-transcription for Feishu users.

Root cause

The Feishu adapter maps audio messages to MessageType.AUDIO via _resolve_media_message_type() and _resolve_normalized_message_type(). This was correct before the Telegram AUDIO/VOICE split, but now AUDIO means "file attachment" (no STT) while VOICE means "voice message" (goes through STT).

Why this fix is minimal

Changes four call sites in gateway/platforms/feishu.py from MessageType.AUDIO to MessageType.VOICE:

• _resolve_media_message_type() — audio/* media type mapping
• _resolve_normalized_message_type() — preferred="audio" resolution
• _should_batch_media_event() — media batching check
• _extract_message_content() — text extraction candidate check

No changes to other platform adapters or the core STT pipeline.

What I tested

• Updated test_extract_audio_message_downloads_and_caches to expect VOICE instead of AUDIO
• Added test_audio_message_maps_to_voice_for_stt as a regression test for Feishu voice messages incorrectly classified as AUDIO (file attachment), skipping STT transcription #30822

What I intentionally did not change

• Telegram AUDIO/VOICE handling
• Core STT pipeline logic
• Other platform adapters (Mattermost, WhatsApp, etc.)

[MKI13]

Hermes autonomous autofix diagnosed the failing Contributor Attribution Check.

Cause: linux2010@github.com (Linux2010) is missing from scripts/release.py AUTHOR_MAP.

I do not have push permission to Linux2010/hermes-agent, so I pushed the one-line fix to MKI13:autofix/pr-30829-author-map and opened a PR against your fork branch:
Linux2010#3

Local verification passed:

• python -m py_compile scripts/release.py
• Contributor attribution check reproduced locally: all contributor emails are mapped.

[alt-glitch]

Duplicate of #29235, which makes the same Feishu AUDIO→VOICE mapping change. See also #29295 and #30174 (both already flagged as duplicates of #29235). Fixes #30822, which is itself a duplicate report of the same bug.

[teknium1]

Thanks for the focused Feishu STT fix. The underlying issue remains present on current main: native Feishu audio resolves to MessageType.AUDIO in plugins/platforms/feishu/adapter.py:3802-3824, while gateway/run.py:10414-10458 sends AUDIO to the attachment-only path and VOICE through transcription.

Problems

• The PR edits gateway/platforms/feishu.py, but that file was deleted when commit 476d8d9cc moved the active adapter to plugins/platforms/feishu/adapter.py. The production edits need to be ported to the active adapter.
• The new test imports gateway.platforms.feishu.FeishuAdapter; current Feishu tests import plugins.platforms.feishu.adapter.FeishuAdapter (for example tests/gateway/test_feishu.py:1477).

Suggested changes

• Salvage the four mapping/batching/extraction edits into plugins/platforms/feishu/adapter.py:3321-3325, :3760-3765, and :3802-3824, and update the test import to the plugin path.

Automated hermes-sweeper review.

[teknium1]

Current main deleted this adapter path in 476d8d9cc; port this resolver change (and the related batching/extraction edits) to plugins/platforms/feishu/adapter.py, the active Feishu implementation.

[teknium1]

The Feishu native-voice classification fix landed in #73515 via @wuli666's #29235 (earliest). Thanks for identifying the same gap!

(Landed via #73515, merge c911a5f10f.) Closing.