michi
道, "the way"
my approach to building software, embodied in an agent-centric toolkit.
01
The Problem
221 tests passed. Two bugs escaped. The agent grades its own homework — and more tests won’t fix that.
02
The Iteration Cycle
Explore → Plan → Execute → Verify → Document. Each cycle builds richer context for the next. Spirals, not circles.
03
Verification
Five layers of verification. Co-designed scenarios. You can only give the agent as much rope as your verification can catch.
04
Nine Skills
Claude Code slash commands for every phase — bootstrap, explore, plan, execute, debrief, sustainability, scenarios, workshop, PR prep.
Explore
Core concepts
The principles and patterns that shape every Michi session.
Reference
The machinery — skills, scenarios, patterns, the docs structure itself.
Start pairedYour first epic should always be Paired.The context that makes Entrusted viable doesn’t exist yet. Start slow, earn speed.
your-project/
# 1. Install the skillscp -r michi/toolkit/skills .claude/skills
# 2. Bootstrap your project/michi-bootstrap
# 3. Plan your first milestone/michi-planning