Support agent

Customer support automation: intent classification → KB retrieval → ticket creation with SLA + escalation lineage.

When to use

You're building a support/help-desk agent that takes inbound messages, either answers them from the knowledge base or files a ticket, and never loses the conversation lineage.

The loop

classify intent → search KB → either answer or file a ticket → log lineage → maybe escalate

What gets generated

File	Role
`AGENTS.md`	Support loop + SLA + escalation rules
`SOUL.md`	"Warm but brief, no corporate cliché" voice
`TOOLS.md`	Ticketing + KB + Slack MCP recommendations + SLA matrix
`MEMORY.md`	Three layers (session conversation / tickets / skill metrics)
`SKILLS/classify-intent/`	question / bug / feature / billing / other
`SKILLS/retrieve-kb-answer/`	KB search with confidence threshold
`SKILLS/file-ticket/`	Lineage-preserving ticket creation
`SKILLS/escalate-if-unresolved/`	SLA-driven escalation with notify-user
`scripts/test_task.sh`	Inbound message replay
`scripts/verify_output.py`	Reply / ticket lineage check

Validators

Validator	What it checks
`structure`	Every blueprint-generated file present and parses
`ticket_lineage`	Every created ticket has `conversation_id` + non-empty `lineage`
`sla`	Every ticket priority is in `{P0, P1, P2, P3}` + `sla_due_at` populated

SLA matrix (also rendered into TOOLS.md)

Priority	First-response SLA	Resolve SLA
P0	15 min	4 hr
P1	1 hr	1 day
P2	4 hr	3 days
P3	1 day	1 week

Memory schemas

conversation — message log with intent + KB consultation history
ticket-history — durable ticket records with required lineage

Hero demo

Zulip + support cookbook shows this blueprint on zulip/zulip.