Guardrails and runbooks

The topic list your agent should decline, and the procedures it should follow.

Guardrails

The Guardrails panel in Agent → Configuration carries one master switch plus a topic list seeded from your industry. On Custom the list is advisory: Humaner returns it, your agent enforces it. Turning the master switch off keeps your selection saved without applying it.

State	Stored selection	Applied
Master switch on	Yes	Yes
Master switch off	Yes	No

Enforce it in your prompt

Every retrieve response carries the list as `forbiddenTopics`, so your prompt stays in sync without polling the dashboard. Add it as a refusal instruction — Humaner cannot enforce it for you because your agent owns the prompt.

Runbooks

Runbooks are approved procedures for recurring requests. Call match-runbook with the customer message; the response names which of your dashboard runbooks matched so your agent follows that procedure instead of improvising.

curl -X POST https://app.humaner.io/api/v1/intelligence/match-runbook \
  -H "Authorization: Bearer hm_live_xxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "agentId": "YOUR_AGENT_PUBLIC_ID",
    "ticketContext": "I was charged twice this month"
  }'

Desk / Runbooks

Runbooks

New runbook

Refund outside window

Approved

Portal login loop

Draft

Trial extension

Auto-generated