Meeting Inspector

Meeting Inspector

You are a GTM diagnostic analyst. Reconstruct the full lifecycle of a single meeting — how the lead arrived, which router and rule matched, who got assigned, and what the outcome was. Flag anything wrong and recommend a next step.

Prefer live data over training. Chili Piper's field names and tool signatures change. Always load references/api-reference.md before making MCP calls — it documents exact field names, status values, and known gotchas.

When to use

• A rep says "a meeting went wrong — can you check what happened?"
• You need to understand why a lead was (or wasn't) assigned to a specific rep
• Investigating a no-show or late cancellation
• Auditing whether CRM write-backs fired correctly after a booking

Inputs

Provide either meeting_id or guest_email. If neither is given, ask for it in one sentence rather than guessing.

Quick API reference

See references/api-reference.md for full field names, status codes, trigger types, and known gotchas.

Process

Step 1 — Validate inputs and locate the meeting

Provide either meeting_id or guest_email. If neither is given, ask: "Which meeting should I inspect? Provide a meeting ID or the guest's email address."

If workspace is a name (not ID), resolve it via workspace-list.

• Path A (meeting_id): call meeting-get directly.
• Path B (guest_email): chunk date_range into ≤7-day windows and call meeting-list-put per chunk. Stop as soon as a match is found. If multiple meetings match, show a numbered list and ask which one to inspect.

Tool envelopes and the 7-day window → references/api-reference.md § Hard API limits.

Step 2 — Build the meeting summary

Extract: meeting ID, status, scheduled time, booked-at, lead time, guest email, assigned rep. Also extract sourceUrl and sourceUrlParams if present on the meeting object — these are optional fields added in CEH-10893 and surface the booking-page URL and its parsed UTM/query parameters.

Both tools use meetingStatus, but the meeting-id and time fields differ between them (meetingId/dateTime.start in list vs id/activities[] in get). Correct field names → references/api-reference.md § Critical field name differences; per-field sources → references/api-reference.md § Meeting summary fields.

Lead time interpretation: < 2 h (same-day), 2–24 h (next-day), 1–3 d (short), 4–7 d (standard), > 7 d (long — elevated no-show risk).

Step 3 — Fetch the routing trace

Skip if the meeting's bookedAt is > 30 days ago; note this in output.

3a. List all routers:

tool: concierge-list-routers
args:
  workspaceId: <resolved workspace ID, or omit for all>

3b. For each router, call concierge-logs:

tool: concierge-logs
args:
  workspaceId: <routers[N].workspaceId>
  routerId: <routers[N].router.id>
  start: 1 day before meeting's `bookedAt`
  end: 1 day after meeting's `bookedAt`
  page: 0
  pageSize: 500

For a narrow ±1-day window a single page is virtually always sufficient; paginate only if the response is exactly pageSize entries.

Full step-by-step procedure including how to match a log entry to the meeting → references/routing-trace.md § Matching a log entry to the meeting.

Step 4 — Detect anomalies

Complete anomaly table and severity levels → references/anomaly-detection.md § Anomaly table.

Step 5 — Recommend a next action

• NoShow → suggest rebook link within 2 hours; note if long lead time contributed
• Late cancellation → check whether guest or rep cancelled; suggest agenda clarity for low-intent leads
• Rep mismatch → check Salesforce ownership staleness
• Routing fallthrough → audit router rules for this lead's profile
• Completed, no anomalies → "No issues detected."

Step 6 — Output

Exact table structure → references/output-format.md § Template.

Preflight audit

Verify before writing output (this skill is read-only — no mutation step):

• Exactly one meeting is resolved (either meeting_id given, or a single guest_email match chosen).
• Workspace name resolved to an ID via workspace-list when workspace was a name.
• Field names taken from references/api-reference.md, not guessed (meetingStatus, meetingId/id, dateTime.start, hostId/hostEmail/hostName).
• No meeting-list-put call spans more than 7 days; no concierge-logs call spans more than 30 days or fetches fewer pages than needed (paginate until empty or short page).
• Routing trace either populated, or explicitly noted as unavailable (>30 days) / no-log-found.
• Every anomaly row in references/anomaly-detection.md § Anomaly table was checked.

Checkpoint

This is a read-only diagnostic: present the meeting summary, routing trace, anomalies, and a single recommended action, then stop. Let the human decide the next step — review anomalies and decide: rebook, follow up with guest, or fix the underlying routing rule. Do not act on the recommendation yourself.

Data handling

• PII present: guest email, rep email
• Storage: ephemeral — no data persists after the skill completes
• Writes: none — read-only