Support reply drafting with escalation

Draft support replies with AI; require human approval for edge cases

For: Support, CX, operations

What it does

Input: an inbound support ticket
AI summarizes the issue
AI drafts a reply
Confidence check — low confidence or refund detected → require approval
Approved reply sent; everything logged

Why it works

High-volume pain point, immediate workload reduction. Shows that Vocion is not just text generation — it handles control flow + human review. Bonus: "Know exactly why the AI responded this way" from the audit trail.

Flow

1Webhook from your help desk → workflow trigger
2summarize_ticket skill runs
3draft_reply skill runs with summary as context
4If confidence < threshold OR keywords match (refund, legal, escalation): pause for approval
5Else: auto-send, still logged

Repo structure

corecontext-support-reply/
  context/
    skills/
      summarize_ticket/
      draft_reply/
    workflows/
      support_triage/
        workflow.yaml
  integrations/
    zendesk-source.ts
    confidence-gate.ts  # plugin: routes high-risk to approval
  README.md

Demo flow

1New ticket arrives via webhook
2Workflow summarizes + drafts reply
3Low-risk → auto-sent + logged
4High-risk → review queue, human edits + approves
5Every reply traceable to its prompt version

View repo Build your own Get started

Status: spec only — repo coming soon. The structure here reflects how it'll be packaged.