Writing workflows

This track teaches how to build ArcFlow workflows beyond a single agent step: linear pipelines, controlled handoffs between steps, graph routing, deterministic testing, and basic retry and timeout configuration. Each lesson is runnable with the built-in stub provider unless noted otherwise.

What you will learn

LessonTopic
01 Linear pipelinesRegister ordered steps and read WorkflowResult
02 Chaining output to inputHow prior step text reaches the next agent
03 Graph workflows introNodes, edges, conditions, and join basics
04 Testing with stub responsesworkflow.test() and stub_responses
05 Retry and timeouts basicsretry(), timeout(), and step_timeout()

Prerequisites

Complete Install and build and the fundamentals track. You should also finish Writing agents so Agent, ContextPolicy, and multi-step pipelines are familiar.

Quick sanity check:

bash
python -c "from arcflow import Agent, Workflow; print('ready')"

How these lessons are structured

Every page follows the same sections: Before you start, Concept, Example, Verify, and Next. Run each example as a standalone script. No API keys are required until you connect a live provider in Python quickstart.

After this track

GoalNext document
Guided trace verificationTrack A: First workflow
Full graph patternsTrack D: Graph workflows
Recovery and resumeRecovery and resume
Tool-calling agentsTools track