Skip to main content
The Documentation tab teaches the method: six steps, capable first, then safe. This tab is where you copy a working agent. Each example below is one agent for a payments operations team, built in the same six steps, with the concrete values filled in: the job, the evaluation categories, the tools and their credential scopes, the runbook skill, the trigger, the seeded cases, and the guardrails added last. Every page reads at its .md URL as a runbook for the runtm-api CLI, because each dashboard step carries its CLI equivalent in an agent-only block (hidden on the website, present at the .md URL).

​
Payments operations

Payment support agent

Reconstructs a charge, decline or refund from Zendesk, Stripe and the ledger and drafts a cited reply. Slack mention and email inbox. Approval before anything reaches the customer.

Fraud and risk review agent

Every 30 minutes, turns new Radar reviews and velocity alerts into a review packet posted to Slack. Enforcement stays human.

Merchant underwriting agent

Compares KYB results, registry data and the application, and writes a decision memo with open items. Linear tickets and a weekly periodic review.

Multi-agent: underwriting re-review

An Underwriting Agent triggered from Slack delegates research to a Risk Agent and a block to an Engineering Agent, each behind a human approval. How subagents are triggered, attributed and graded.
The three single-agent examples follow the same order as Build an agent; the multi-agent example builds each agent that way and then wires them together. If you are building something else, pick the closest one and change the values.

​
Recipes

One task each, five minutes each.

Connect Stripe with a restricted read-only key

Stripe as a custom provider, the restricted key, and the skill that requires it.

Add a KYB provider as a custom tool provider

Middesk, Persona or Alloy as a static-key provider scoped to one agent.

Add an MCP server with an API-token connection

A stdio MCP server whose secret lives on a connection, never in the snapshot.

Route one Linear team to a different template

Team overrides on the agent’s Linear trigger.

Test a scheduled agent before enabling it

Create disabled, run now, read the session, then enable.

Request an approval from inside a run

The runtm-approval helper and who can resolve it.

Delegate work to another agent

The delegate script, the gate hook and the two-line command the dashboard recognises.

​
Looking for API patterns?

Authentication, sessions at scale, streaming prompts over WebSockets, session files, long-running prompts, activity and telemetry, and outbound webhooks moved to the API Reference tab under Patterns.