Cheatsheet

A quick reference for someone who knows Michi and wants a fast reminder. New here? Start with the How-To Guides reading guide.

The iteration cycle

Every piece of Michi work — except one-time bootstrap — runs the same six phases:

Phase	What it is
Explore	Gather context. Read the code, the docs, the problem space.
Brainstorm	Shape the approach. Decide what before how.
Plan	Write it down. Scope, acceptance criteria, verification — the contract.
Execute	Do the work. Test after every change; log decisions as you make them.
Verify	Check it. Run the scenarios — not just “tests pass.”
Document	Capture what was learned. This is what makes the next cycle better.

It is fractal — an epic runs the cycle, each milestone runs it, a bug fix runs it — and it spirals: every pass produces a deliverable and richer context for the next pass.

The use cases → guides

Your situation	Guide	Carried by
Set a project up for Michi	Bootstrap Michi	/michi-bootstrap
Investigate / evaluate / think through	Research and Explore	/michi-explore
One bounded change	Implement a Task or Story	/michi-workshop
A multi-phase, multi-milestone effort	Run an Epic or Project	/michi-planning → /michi-session → /michi-debrief
Fix something broken	Troubleshoot and Fix a “Thing”	/michi-workshop bugfix
Get better at working with agents	Improve Your Agent Practice	export + co-read transcripts

The two modes

Mode	What	When
Paired	Human present, tight loop, reviewing each step	New codebase, uncertain work, the first epic — always
Entrusted	Agent has wider initiative; human reviews at gates	Earned, once verification and shared context support it

Moving back to Paired isn’t regression — it’s knowing when to slow down. See Modes.

Core disciplines

These hold regardless of use case:

• Surface assumptions — say what you’re assuming before you act. The cheapest bug to catch.
• Plan as contract — the plan doc, not conversation memory, is the source of truth. It survives context compaction.
• Verify before “done” — “tests pass” is necessary, not sufficient. Run the scenarios. “I believe it passes” is not evidence.
• Capture decisions — log them as you make them. Retroactive recall is unreliable.
• Match ceremony to the work — a bug fix doesn’t need an epic’s process. A workshop is about less ceremony, never less rigor.

When to escalate

A workshop that grows a design decision, a multi-system change, or a scope expansion is no longer a workshop. Stop and switch to /michi-planning. Escalating is the correct move, not a failure.