Greptile Summary

This PR fixes two Feishu-specific edge cases to prevent internal path leakage: it removes Feishu from the VOICE_BUBBLE_CHANNELS set (so TTS audio is delivered as MP3 rather than triggering voice-bubble auto-routing), and it prevents local file paths from being surfaced to end users when media upload fails (replacing the leaked path with a user-friendly Chinese error message for local paths, while keeping the remote URL fallback intact).

Key changes:

• src/tts/tts.ts: feishu removed from VOICE_BUBBLE_CHANNELS, so resolveOutputFormat returns MP3 defaults instead of opus for Feishu TTS output; audioAsVoice will no longer be set for Feishu replies.
• extensions/feishu/src/outbound.ts: Added isLocalMediaPath helper (checks path.isAbsolute or file:// prefix); sendText catch block now returns early with "图片发送失败，请重试" instead of falling through to send the raw local image path; sendMedia fallback suppresses local paths with "音频发送失败，请重试" while still forwarding remote URLs as a 📎 link.
• Tests comprehensively cover absolute path, file:// URL, and remote URL cases for sendMedia, and the sendText image path leakage scenario.

One observation worth noting: The error messages ("图片发送失败，请重试", "音频发送失败，请重试") are hardcoded in Simplified Chinese with no i18n/l10n hook. This is consistent with Feishu's primary user base, but may be worth revisiting if multi-locale Feishu deployments are expected.

Confidence Score: 4/5

• This PR is safe to merge — changes are small, focused, and well-tested with no regressions introduced.
• All four changed files have clear, targeted logic with corresponding tests that pass. The isLocalMediaPath helper correctly covers both absolute path and file:// URL cases. The only minor concern is hardcoded Chinese-language error strings with no i18n hook, which could be a problem in multi-locale deployments but is not a blocking issue for the stated goal of this PR.
• No files require special attention.

_{Last reviewed commit: 10c2e7f}

This pull request has been automatically marked as stale due to inactivity.
Please add updates or it will be closed.

Closing this PR rather than keeping it stale.

Why not land:

• The core TTS change conflicts with current main. Feishu is now intentionally treated as a native voice-note channel: extensions/speech-core/src/tts.ts:588 includes feishu, and extensions/speech-core/src/tts.ts:810 requests the voice-note target. Removing Feishu from the voice-note path would regress the behavior tracked in [Bug]: TTS audio sent as file attachment instead of voice message in Feishu #59588 / [Feature]: 1: Feishu voice bubble — MP3/WAV sent as file instead of voice bubble、2: Per-agent TTS voice override for Microsoft Edge TTS #56231 / [Bug] TTS sends duplicate text + audio, and format not configurable per channel #70092.
• Current Feishu media routing already sends .opus/.ogg and audio/ogg/audio/opus as native msg_type=audio at extensions/feishu/src/media.ts:539.
• Current Feishu media loading also passes mediaLocalRoots into the runtime loader at extensions/feishu/src/media.ts:613, so the old broad local-path fallback problem is not the same shape anymore.

What may still be worth splitting:

• A tiny Feishu-only PR to avoid leaking a plain local image path if sendText auto-converts a local image path and the upload fails. Current code still falls through to text after sendMediaFeishu throws at extensions/feishu/src/outbound.ts:184. That should be handled without changing Feishu voice-note/TTS routing.

Local proof on current main: pnpm test extensions/feishu/src/media.test.ts extensions/speech-core/src/tts.test.ts passed (2 files, 33 tests).