Track Weixin native voice-bubble delivery for Hermes hosted audio replies

[Sapientropic]

Summary

Track whether Codeksei + Hermes can eventually deliver Weixin audio replies as native voice bubbles instead of ordinary file attachments.

This is not just a transport detail. For companion-style interactions, a Weixin voice bubble feels much more alive and conversational; a downloadable file attachment is usable, but it has a colder, more mechanical UX.

Current status as of 2026-04-21

Native outbound Weixin voice bubbles should be treated as not solved.

The strongest public repro is still:

• text delivery works
• file attachment delivery works
• real .silk conversion works
• sending the resulting .silk through outbound voice_item / native voice payload is rejected or dropped

Specific evidence:

• NousResearch/hermes-agent#9971 remains open. The 2026-04-16 repro reports that real .silk + voice_item still returns {"ret": -2} from ilink/bot/sendmessage across multiple payload variants.
• NousResearch/hermes-agent#11571 was merged as a partial fix, but it deliberately routes Weixin send_voice() to file attachments. The PR text says full voice-bubble TTS remains open.
• Tencent/openclaw-weixin#18 remains open and includes evidence that a local outbound VOICE send path can report success while the WeChat client receives no voice bubble.
• Tencent/openclaw-weixin#9 adds native outbound voice support, but it remains open and unmerged.
• @tencent-weixin/openclaw-weixin@2.1.9 still does not expose a merged outbound sendVoiceMessageWeixin path in the published tarball. Its public README still describes sendMessage as text/image/video/file, and getUploadUrl as media types 1=IMAGE, 2=VIDEO, 3=FILE.

Why this matters for Codeksei

Codeksei's hosted Hermes flow can currently rely on audio file fallback for delivery, but that is not the desired companion UX.

For ADHD companion / check-in interactions, native voice bubbles are materially better because they:

• feel like a normal Weixin conversation instead of a file transfer
• reduce friction on mobile
• make proactive or reassuring check-ins feel more human
• better match the mental model of short spoken replies

So this deserves a dedicated tracking issue even if the current fallback is technically functional.

Current recommended behavior

Until upstream evidence changes:

• Do not claim Weixin native outbound voice bubbles are supported.
• Prefer reliable file attachment fallback for audio replies.
• If Codeksei or Hermes has a native voice_item code path, keep it behind an explicit experiment / probe rather than enabling it by default.
• Treat ret=-2 on sendmessage as a real delivery failure, not success.
• Keep the distinction clear in docs and smoke output: “audio attachment works” is not the same as “native voice bubble works”.

Things to re-check before closing

Close this only after one of these becomes true:

1. Tencent documents outbound voice_item support for personal Weixin iLink bots, including the exact required upload/send fields; and a real-device test confirms delivery as a native voice bubble.
2. Tencent/openclaw-weixin merges and releases a working outbound voice path, and Codeksei/Hermes verifies it on current stable.
3. Hermes adds a verified native Weixin voice-bubble path with real-device proof, not just unit tests that construct voice_item JSON.
4. Tencent or maintainers explicitly confirm outbound native voice bubbles are unsupported for this bot API, and Codeksei records the fallback as a product limitation rather than a pending bug.

Follow-up checklist

• Re-check NousResearch/hermes-agent#9971 for maintainer response or closure.
• Re-check NousResearch/hermes-agent#11571 and any follow-up PRs after the file-attachment fallback.
• Re-check Tencent/openclaw-weixin#18 and Tencent/openclaw-weixin#9.
• Re-check the latest @tencent-weixin/openclaw-weixin npm tarball for a real sendVoiceMessageWeixin implementation.
• If a native path appears, run a real-device smoke that confirms the Weixin client displays a voice bubble, not a file attachment.
• Update Codeksei / Hermes hosted docs to say clearly which behavior is available.

Related tracking

• Codeksei broader Tencent plugin / overseas-login tracker: Track Tencent official Weixin plugin alignment and international WeChat login limits #8
• Hermes voice-bubble issue: [Weixin] TTS audio messages are not delivered as voice bubbles — need OGG → SILK transcoding NousResearch/hermes-agent#9971
• Hermes fallback PR: fix(weixin): Batch-3 salvage — voice fallback, MEDIA extraction, native markdown rendering NousResearch/hermes-agent#11571
• Tencent plugin voice issue: [Feature Request] Support sending voice messages (VOICE type) to WeChat users Tencent/openclaw-weixin#18
• Tencent plugin native voice PR: Add outbound Weixin native voice-message support Tencent/openclaw-weixin#9
• Community CDN voice-flow experiment: feat(weixin): implement Weixin CDN voice upload & send flow jazzqi/openclaw-mimo-tts#4