Teams In-Meeting Side Panel Incident Copilot Prompt
Build an in-meeting Teams app (side panel + shared meeting stage) that acts as a live incident copilot — tracking timeline, action items, and current owner while a war-room call is in progress.
- Target user
- App developers building real-time incident tooling for Teams meetings
- Difficulty
- Advanced
- Tools
- Claude, ChatGPT
The prompt
You are a senior Teams app developer who has shipped in-meeting apps that on-call teams actually trust during live incidents. I will provide: - The incident data model (timeline events, roles, severity, comms) - The source of truth (incident API, PagerDuty, internal tool) - Auth model (SSO, on-behalf-of to a backend API) - Who is in the meeting (responders, exec observers) Your job: 1. **Surface design** — map each meeting surface to a job: the side panel (private, per-attendee triage feed and quick actions), the shared meeting stage (one synchronized timeline everyone sees), pre/post-meeting tabs (briefing and auto-postmortem). Explain `frameContext` detection via the Teams JS SDK to render the right UI per surface. 2. **Manifest wiring** — produce the manifest entries: `configurableTabs` with `meetingSidePanel` and `meetingStage` scopes, `meetingSurfaces`, and the `supportsAnonymousAccess` decision for guest responders. 3. **Live sync** — design the real-time channel (SignalR / WebSocket) so a timeline update from one responder appears on the shared stage within ~1s; handle reconnect and late-joiner state hydration. 4. **Roles & RBAC** — use `getMeetingDetails` and the participant role (Organizer/Presenter/Attendee) to gate destructive actions (declare resolved, page next tier) to incident commanders only. 5. **Quick actions** — side-panel buttons that, via SSO on-behalf-of, write to the incident API: assign owner, add timeline note, bump severity, request a page. Each must be idempotent and confirm inline. 6. **Auto-postmortem handoff** — on meeting end (`meetingEnd` event / Graph subscription), snapshot the timeline into a draft postmortem and post a card to the incident channel. 7. **Failure modes** — what the panel shows when the backend is down, when SSO token exchange fails, and when a guest lacks permission. Output as: (a) the manifest fragment, (b) a `frameContext` + role-resolution flow, (c) the live-sync architecture, (d) the quick-action API contract, (e) the meeting-end → postmortem sequence, (f) an offline/degraded UX spec. Bias toward: one shared source of truth, role-gated destructive actions, graceful degradation when the backend blips mid-incident.