Stage Manager

When the renewal goes quiet, page the owner — not the customer.

Every company pages an engineer when a server goes down. Almost nobody pages the human who owns the account. PagerDuty phones you to acknowledge an alert. Ack is not a decision — we capture one in about sixty seconds.

Open the skill Judge quick start

Scroll for the cue sheet

Slack can wait. This account can’t.

Support loops. SLA turns red. An agent hits needs_human. Health stalls before renewal. Chat notifications hide. Phone doesn’t — if the call is rare, short, and for the right person. We enforce a per-owner call budget so “rare” is policy, not a hope.

One engine. Four cues.

  1. Cue arrives (fixture, stdin, or POST /cue)
  2. Call sheet — owned? severity? quiet hours? owner budget?
  3. Dress rehearsal (dry-run) by default — or curtain up (live call)
  4. Identity — spoken stage code (binds the decision to the call-sheet owner)
  5. Line reading — a 1 / 2 / 3 decision menu on the CS phone
  6. Prompt book (audit log) + action intent (system handoff)

stuck_support

Take over in chat now · Assign to SE / specialist · Not now — snooze 2 hours

agent_needs_decision

Approve A — allow the one-time exception · Approve B — require the signed amendment first · Reject — escalate to manager

sla_risk

I own this ticket now · Page backup CS/SE · Acknowledge risk, no action

health_onboarding

Book SE technical session · Mark watchlist — no outreach yet · Flag churn risk for manager

How judges run this

Clone → dress rehearsal (no key) → action-intent dry-run → optional curtain-up.

git clone https://github.com/assafbar2/customer-success-voice-signal-hackathon.git
cd customer-success-voice-signal-hackathon/skills/customer-success-voice-signal
npm install

# Dress rehearsal — default, no ring
npm run signal -- --fixture stuck_support_acme.json
npm run signal -- --stdin < events/webhook_stuck_support.json

# Or: HTTP inbound (same engine)
npm run serve-cue &
curl -sS -X POST http://127.0.0.1:8787/cue \
  -H 'content-type: application/json' \
  -d @events/webhook_stuck_support.json

# Decision → system handoff (Slack/GitHub live when env set)
npm run apply-action -- --last --dry-run
npm run apply-action -- --last --adapter slack

# Curtain up — real CALL-E (operator only)
# Key: https://dashboard.heycall-e.com/account/api-keys
# cp .env.example .env  → CALLE_API_KEY + CS_OWNER_E164
npm run signal -- --fixture stuck_support_acme.json --live PLACES

Exit codes: 0 ok · 2 HOLD · 3 failure. Action intents land in data/actions/pending/. Never commit .env.

Operator-proven live ladder

Real curtain-up calls against CALL-E — logged honestly, misses included.

  • stuck_support — voicemail → no_answer, then curtain-up → line reading 1 (take over in chat)
  • agent_needs_decision — earlier unclear captures, then curtain-up → line reading 1 (approve A)

Callee is always the CS owner. Customer never gets this call. Full probe table: research/calle-api-notes.md. Redacted rows: submission/evidence.