Skill Details
Runtime
(claude-4-6-sonnet)
Example user message
[userMessage]: Design an n8n workflow that classifies incoming support tickets, retrieves account context, drafts a response, routes high-risk cases to a human, updates Zendesk, and recovers safely when the CRM, model, or ticket API fails.
Example agent response
# AI Agent Workflow Architecture & Recovery Plan — Support Escalation
## Decision summary
Use event-driven n8n orchestration with deterministic validation before model classification. Keep high-risk responses and every consequential account action behind human approval.
## Target flow
Zendesk webhook → schema validation and deduplication → account-context retrieval → bounded classifier → policy gate → human approval when required → Zendesk draft/update → audit event.
## Recovery controls
| Failure | Containment | Recovery |
|---|---|---|
| CRM timeout | Stop context-dependent automation after two bounded retries | Create a human-review task with the ticket payload and correlation ID |
| Invalid model output | Reject the response against the JSON schema | Re-run once with repair instructions, then route to review |
| Zendesk update timeout | Preserve the idempotency key and checkpoint | Verify ticket state before replay to prevent duplicate updates |
Launch only after duplicate, timeout, authorization, provider-outage and replay tests pass with retained evidence.
By purchasing this skill, you agree to our terms of service
CLAUDE-4-6-SONNET
Turns a business process or unreliable AI agent into an implementation-ready workflow architecture. Maps agents, tools, APIs, data contracts and human approvals; defines retries, timeouts, idempotency, fallbacks, dead-letter handling, observability, cost controls, tests, phased rollout and a recovery runbook. Works with n8n, Make, Zapier or custom stacks without inventing integrations or readiness claims.
...more
Added 6 days ago
