You are a senior automation engineer. Build the following optimized workflow.

# Triage and draft replies for inbound support tickets

Spion recorded this as a manual browser process (6 steps, ~5 min per ticket, high volume) and redesigned it as a consolidated automation (4 steps). Build the OPTIMIZED workflow below, not a replay of the manual clicks.

Overview: On a new customer ticket, classify it, set priority, draft a reply grounded in the help center and saved macros, and route it to the right queue — so agents review and send instead of starting from a blank box.
Tools: Zendesk (or Intercom), Claude, Slack
Trigger: Helpdesk event (When a new ticket is created).
Efficiency target: 6 manual steps down to 4 (33% fewer), saving ~5 min per ticket and cutting first-response time.

## Optimized workflow steps

1. [Zendesk] Read the ticket
   What: Read the subject, body, customer, plan, and recent ticket history.
2. [Claude] Classify, prioritize, and draft (added by Spion: net-new value)
   What: Categorize (billing, bug, how-to, outage, cancellation); set priority from impact and plan; draft a reply grounded in the matching help-center article or macro, with citations; detect sentiment.
   Why: turns a cold ticket into a routed, near-ready response while keeping a human in the loop.
3. [Zendesk] Tag and route
   What: Apply the category, priority, and tags; assign to the right queue; attach the draft as an internal note (not auto-sent).
4. [Slack] Escalate the urgent
   What: Post any outage, churn-risk, or angry-sentiment ticket to #support-escalations for a human to grab now.

## Original manual process (reference only, do not replicate)
1. Open each new ticket
2. Read and categorize it
3. Set a priority
4. Search the help center / macros
5. Write a reply
6. Assign it to a queue

## What I need from you
1. Build the optimized workflow exactly as designed above: production-ready code (Python or JavaScript) or a platform recipe (Workato, Make, Zapier, or n8n), whichever fits these tools best.
2. Preserve the consolidation. Do not expand it back into the manual click-by-click process.
3. Recommend the right trigger and cadence (schedule, webhook, or event). If it recurs, state the exact schedule.
4. Define the data contract for each step: the exact input fields it reads, the output fields it produces, and how fields map between tools.
5. Specify every authentication requirement: which services need OAuth, API keys, or service accounts, and the exact scopes.
6. Draft replies as internal notes for an agent to review and send — never auto-reply to the customer — and ground every answer in the help center or an approved macro rather than inventing policy.
7. Add robust error handling (missing body, auth failures, empty results, rate limits with retries) and make the workflow idempotent so it is safe to re-run.
8. Include a short test plan: a sample ticket and the expected classification + draft, so I can verify it before going live.
9. Finish with exact, copy-and-paste setup instructions so I can run it today.