B.O.S.S.

Guide

Using the team, rung by rung.

Each mode is additive — unlocking the next never removes what you had, and boss unlock <mode> is always your call. You don't memorize any of this: boss map lists what your project has and what's one unlock away.

the one idea

When to use BOSS, and when to just talk to Claude

BOSS doesn't sit between you and Claude. You build exactly as you do now. The split is simple:

  • “How do I build this?” — just ask Claude. No skill, no ceremony. That's most of your day, and BOSS stays out of it.
  • “Should I build this? Is it working? What did I decide?” — that's a BOSS verb. Those are the seams between building, where things get lost.

While you're heads-down, BOSS does almost nothing. That's the design, not a gap.

the walk
  1. Quickstart 3 on the team 16 skills
  2. MVP 10 on the team 45 skills
  3. V1 10 on the team 47 skills
  4. Scale 11 on the team 48 skills

You unlock each stop. Nothing is ever removed — the roster only grows.

Quickstart — get the idea out of your head

Capture freely with /triage. Pressure-test one with /canvas, filled a few cells at a time rather than all at once. You never pick a framework — you answer questions, and the frame is a view you switch. The same answers render as the Humane Product Canvas (the default), a Lean Canvas, or a Business Model Canvas; /canvas --frame lean re-renders what you've already said rather than interviewing you again, and a cell you haven't answered shows as _(not yet)_ in every frame instead of quietly disappearing to make a view look complete.

Commercial prompts are folded into every cell (who exactly has the problem, what they use instead, who pays and why). And two questions ride along in all three frames, not just the humane one: who this could harm or exclude, and the principles you'll hold when holding them costs you something. Un-defaulting the framework was never meant to make those optional.

Work out who it's for with /persona — your target user as a voice you can actually consult, both to decide what to build and to react to what you built. It carries a visible ledger of how much of that user is real evidence and how much is still a guess, and it starts at all guess. It's a way to sharpen the questions you take to a real person, never a way to skip them. Record what you learn with /evidence, and what you chose with /decide. A project that lives here forever is a legitimate project.

MVP — build the first working version

Adds the spec discipline (/spec, with validated-learning and evals fields), the smoke gate, and the demand test — /pretotype, which builds a fake door before you build the real thing. Design tokens get scaffolded here too, the moment there's real UI, so style never gets locked into code, and the designer arrives with them — one designer for both halves, what it looks like and what it does (/design-review, /ux-check).

It's also where the questions start pointing outward. /comp-eval goes and finds out who else is solving this — features, pricing, and the honest “why they might win” — then keeps it as a living set you add names to, rather than a slide you drew once. /roadmap weighs what to build next into a short bet-list with a mandatory no-list, because a roadmap that only says yes isn't a decision.

V1 — make it shippable

V1 hires nobody. It's the one rung that adds no agents at all, and that's deliberate — by now you have the team; what you don't have is a way to see what you've already made. So it adds two views. /design-library generates a gallery of your components — every variant, all five states — from your code, plus a reuse index the agent reads before writing a new one. /board sequences work across features instead of within one.

Scale — grow it

Adds /incident. Deliberately thin: most projects should never need this rung.

after you ship

The post-launch arc

A second set of verbs whose moment only arrives once something is live — retention, pricing, measurement, health, the first dollar. They install with MVP but stay folded away until you've actually shipped a feature, so a founder with one idea isn't read a menu of nine verbs about pricing.

And a project can end at any rung. /sunset exists because quitting well is a skill, and a tool that only knows how to grow things is lying to you about how this usually goes.

with a cofounder

Building with someone else

AI made each of you faster — which, left alone, makes a two-person team drift apart. You each work in your own AI session, ship in parallel, and quietly stop deciding things together. The uncomfortable part: that erosion is invisible from the inside. Teams that have already drifted still rate their teamwork as fine.

boss team add @their-handle lights up a shared place to decide (/decide), a shared craft commons (/practice), and mentor-cofounder — the one mentor that coaches the relationship: dividing work across different skill sets, decision rights, and the hard conversations about roles, pace and equity. It serves the partnership as a unit and never takes a side.

All of it is dormant when you're solo. Keyed to GitHub handles — no accounts, no seats, no server, no hierarchy.

The full teams guide →

worth remembering

Two things

  • boss status is the one command to remember. Come back after a week and it tells you where you are and what's next.
  • The conscience is always overridable. boss conscience pause --for 8h for a sprint, or mute <moment> to turn down one nudge while the rest keep speaking. It never blocks.

The full guide → · every skill →