Before you build anything, you map the workflow honestly: every step, who owns it, which tool does the work, and where it stalls or waits. The messy version is the useful one.

• How to map a workflow as steps, owners, and tools
• Where work stalls, waits, or gets handed off with no clear quality bar
• Outcome: An honest map of the workflow you are about to build

You draw the line on every step: which ones AI can own end to end, which a human has to keep, and what each AI step must produce before the next one runs.

• How to decide which steps AI owns and which stay human
• What each AI step must output before the next step can trust it
• Outcome: A redesigned workflow with AI and human steps marked

You build the AI steps as skill files in Claude Code, one per step, each saving its own output so you can see exactly where a run went wrong.

• How to write a Claude Code skill that mirrors how you do the step
• Why each step saves its own output file before the next one runs
• Outcome: Working skill files for the AI steps in your workflow

You chain your skills with a single command so raw research data goes in and test ideas come out, running end to end without babysitting each step.

• How one command chains skills into a workflow that runs end to end
• How to turn messy research inputs into testable ideas
• Outcome: A research-to-test-ideas workflow that runs on one command

You push the finished workflow to a GitHub repo you own, then learn the pattern well enough to rebuild any other workflow the same way after the session.

• How to store and version a working workflow in your own repo
• The repeatable pattern for rebuilding any workflow as AI plus skills
• Outcome: Your skills, command, and workflow live in a GitHub repo, plus a method you can reuse