anthropics--knowledge-work-plugins/partner-built/zoom-plugin/skills/rtms/RUNBOOK.md

RTMS 5-Minute Preflight Runbook

Use this before deep debugging. It catches the highest-frequency RTMS issues fast.

Skill Doc Standard Note

• Agent-skill standard entrypoint is SKILL.md.
• This runbook is an operational convention (recommended), not a required skill file.
• SKILL.md is also a navigation convention for larger skill docs.

1) Confirm Architecture Assumption

• RTMS is backend-first media ingestion.
• Frontend is optional and should consume backend outputs (WebSocket/SSE/etc).

If implementation assumes frontend-only RTMS behavior, redesign first.

2) Confirm Event-Triggered Kickoff

• Processing starts only after RTMS lifecycle start events:
  • meeting.rtms_started
  • webinar.rtms_started
  • session.rtms_started
• Stop events should deactivate pipeline.

If media handling starts before lifecycle start, session gating is wrong.

3) Confirm Product-Specific IDs

• Meetings/Webinars: use meeting_uuid
• Video SDK: use session_id
• Use rtms_stream_id from payload for stream context

Using wrong ID field commonly breaks handshake/signature.

4) Confirm Webhook Handling Pattern

• Respond 200 immediately.
• Do heavy work asynchronously.
• Verify webhook signature if secret token is configured.

Slow webhook responses can trigger retries and duplicate stream attempts.

5) Confirm Connection and Heartbeat

• Track one active connection per stream/session reference.
• Handle heartbeat ping/pong per protocol.
• Implement reconnection strategy explicitly.

No heartbeat handling means unexpected disconnects.

6) Confirm Media Subscription/Gating

• Ensure requested media types match your processing path.
• Reject/ignore media packets for inactive sessions.
• Expose pipeline status endpoint for observability.

This avoids silent packet handling when lifecycle is not active.

7) Quick Probe Checklist

• GET /api/health returns service alive.
• GET /api/pipeline/status shows expected active session count.
• Mock/media probes show:
  • media before start -> rejected
  • start event -> pipeline active
  • media after start -> accepted

Copy/Paste Validation Commands

curl -sS "$RTMS_BASE_URL/api/health"
curl -sS "$RTMS_BASE_URL/api/pipeline/status"

Expected: healthy service JSON and correct active pipeline visibility.

8) Fast Decision Tree

• No media at all -> lifecycle event not received or wrong webhook route.
• Duplicate streams -> delayed webhook response or no active-session guard.
• Handshake/auth errors -> wrong credential pair or wrong session ID field.
• Frontend appears idle -> backend bridge not connected, not an RTMS source issue.