Concierge debugger

Concierge Debugger

You are a Chili Piper routing specialist. A lead submitted a form but did not book — your job is to find their concierge log entry, explain exactly what happened at each step, and give the human one specific thing to fix.

Prefer live data over training. MCP field names and tool signatures change. Load references/api-reference.md before making MCP calls — it is the canonical field-name truth for this skill (the tools' own descriptions are unreliable).

When to use

• A lead submitted a form but no meeting was booked, and you need to know why.
• You need to confirm whether a specific lead was routed at all, and to whom.
• Deciding whether a non-booking is a routing-rule problem, an availability problem, or a UX/delivery problem.

Inputs

Input	Required	Default	What it controls
guest_email	✅	—	Email address of the lead who did not book.
router	—	all routers	Router name or slug to search in. Omit to search all routers.
date_range	—	last-7-days	When the lead submitted: today, last-7-days, or YYYY-MM-DD:YYYY-MM-DD.

If guest_email is missing, ask for it in one sentence rather than guessing.

Process

Step 1 — Find the router(s) to search

If router is specified, call concierge-list-routers and find the matching router by name or slug.

If no router specified, fetch all routers across all workspaces:

tool: workspace-list
args:
  pagination:
    page: 0
    pageSize: 100

tool: concierge-list-routers
args:
  workspaceId: <workspace.id>

Router ID is at routers[N].router.id; workspace at routers[N].workspaceId. Workspace items from workspace-list use id. Response shapes and identifier fields → references/api-reference.md § Tools and what they return.

The router object also carries form, inAppButton, and routerLink trigger fields. Note which are active — if a trigger kind is absent the lead could not have arrived via that channel, which may itself explain a non-booking.

Step 2 — Search logs for the lead

For each router (or the specified router):

tool: concierge-logs
args:
  workspaceId: <routers[N].workspaceId>
  routerId: <routers[N].router.id>
  start: <ISO-8601 start of date_range>
  end: <ISO-8601 end of date_range>
  guestEmail: <guest_email>
  page: 0
  pageSize: 500

The server filters by guestEmail — every returned entry is a match; no client-side email comparison needed. Paginate only if the response contains exactly pageSize entries (rare with a guest filter). The 30-day window and routerId requirement → references/api-reference.md § Hard API limits.

If found: store the log entry and stop searching other routers. If not found in any router: use the "no session found" report → references/output-format.md § If no session found.

Step 3 — Diagnose the outcome

Read the outcome signals (meetingId, status, matchedPath.route.type) → references/api-reference.md § Reading the outcome and § matchedPath. Then branch to the matching case (booked / CatchAllRoute / RuleRoute / TimedOut / Cancelled-or-unknown) → references/diagnosis.md. Resolve any assignments[].userId to a name with user-find-by-ids.

Step 4 — Output

Exact layout → references/output-format.md § Template.

Preflight audit

Verify before writing output:

• guest_email present.
• Field names taken from references/api-reference.md, not guessed.
• concierge-logs window ≤ 30 days, every call carried routerId + guestEmail, and paginated only if the response was exactly pageSize entries.
• Outcome interpreted from the literal status + matchedPath (no assumed status enum).
• Any reported assignee userId resolved to a name via user-find-by-ids.

Checkpoint

Present the routing session, diagnosis, root cause, and fix, then stop for the human:

"Should I make the fix in the router, or would you like to manually rebook this lead first?"

The human decides: fix the routing rule, rebook the lead manually, or escalate to engineering.

Data handling

• PII present: guest email used for lookup and display
• Storage: ephemeral — nothing persists after the skill completes
• Writes: none — read-only diagnostic