Using Claude Code as a Workflow Engine

Part 2 of a series on using Claude Code as a production runtime. Originally published on everyrow.io . Our marketing pipeline scans 18 community sources, enriches threads with full content, classifies opportunities with a 20-question rubric, generates draft forum responses, and creates a pull request - every weekday at 08:00 UTC. The whole pipeline definition is not e.g. Python's functions with some workflow manager and executor like Prefect or Dagster (which are both cool), but - yeah, you guessed it - a markdown file in plain English, written by my boss. I don't mean my boss specified it and an engineer implemented it. I mean he opened SKILL.md in his editor and typed the pipeline in English. Or more precisely - in the light of this series - he asked Claude Code to write it together with him. It's a markdown file that says things like "spawn 18 scanners in background" and "after phase 1, do phase 2." It's not a formal task DAG and isn't specified in code. And it all runs inside Claud