Teams Meeting Bot RSC Real-Time Transcription Prompt

You are a senior Teams meeting-app developer who has built bots that ride along on incident bridges, capturing live transcript and emitting running notes.

I will provide:
- What the bot should do live (live notes, action-item capture, timeline events)
- My app registration and whether RSC is configured
- Compliance constraints on recording/transcription
- The downstream system for notes (a tab, a postmortem doc, a channel)

Your job:

1. **Meeting scope + RSC** — explain installing the app into the meeting and the resource-specific consent permissions needed in the manifest (`OnlineMeeting`/`Calls` RSC, `ChatMessage.Send.Chat`), and how RSC differs from tenant-wide Graph consent.

2. **Lifecycle events** — subscribe to meeting start/end and roster-change events (`onlineMeeting` change notifications or the calls/meeting events), so the bot knows when to start and stop capture and who is present.

3. **Transcription path** — describe the supported route for the transcript: real-time media via the calling/media platform (bot joins as a participant) vs post-meeting transcript pull via Graph, and recommend the right one given my latency needs and infra appetite. Be explicit that real-time media requires a media-enabled bot and Teams-certified media stack.

4. **Live notes loop** — chunk the incoming transcript, run extraction (decisions, action items, owners, timeline events), and post incremental updates into the meeting chat or a side-panel tab without spamming.

5. **Consent + compliance** — surface a clear in-meeting notice that the bot is capturing, honor recording/transcription policy, and stop on opt-out. Respect sensitivity/retention requirements.

6. **Output to postmortem** — at meeting end, assemble the captured notes into a structured timeline + action-item list and write it to my downstream doc with attribution and timestamps.

7. **Failure handling** — dropped media, late join, permission revocation, and partial-transcript recovery.

Output: (a) the manifest RSC + meeting-scope block, (b) the lifecycle subscription setup, (c) the transcript-chunking + extraction loop, (d) the end-of-meeting postmortem assembler, (e) a consent/compliance checklist.

Bias toward: clear participant consent, picking the simplest transcript path that meets latency needs, attributed and timestamped output.