For the brain that runs hot, jumps sideways, and somehow ships anyway. Pick your state, pick your time. The recipe writes itself.
Most Claude Code guides are written for people who already know what they want to do. That's not the bottleneck.
The bottleneck is the half-second between "I should work on something" and "I have no idea which thing, also I can feel my brain trying to leave." That half-second is where projects die.
This guide skips the manual. It assumes you already know what Claude Code is. What it gives you instead is a way to point your brain — whatever state it's in right now — at the right move, the right model, the right opening prompt. Pick the state. Pick the time. Get the recipe. Run it.
A new idea hit. You're hot. Energy is the fuel. Direction is the question.
Coming back after time away. Can't remember where you left T-BOT, or what state designdna's in.
Been hammering one thing too long. Brain stale. The bug isn't getting smaller — you are.
Eight half-things on the friction list. Eight choices. Zero forward motion. The thing you're avoiding is here somewhere.
These aren't motivational concepts. They're the specific failure modes that have stalled your work before. Claude Code only helps if it shows up at the right moment with the right move.
An idea fires. You don't write it down fast enough. By the time you find the right file, the spark is gone. Three days later you remember it was "something about ICC and weekly cycles" but the specifics are dust.
The move: Apple Notes inbox + Ferryman's 6am drain. If Claude Code is open, paste it in and say "capture this as a new idea spark, then come back to what we were doing."
Eight things ready to ship. Three live deploys. The Fiverr gig is 35+ days late but you don't even know what's blocking it anymore. Choosing what to touch feels like deciding what to eat when you're hungry — impossible.
The move: Use the picker above. State = Drifting + 15 min. Don't try to solve the meta-problem of "what should I work on" — just commit to a small win.
You return to T-BOT after four days. Can't remember which agent file you were rewriting, what the last bug looked like, whether the paper-trade loop ran clean. Twenty minutes of staring before any work happens.
The move: End every session with a 3-line update to project CLAUDE.md: "active focus: X. blocked on: Y. next session: Z." Claude can write this for you — ask before you close.
Someone shipped a thing you were considering. Suddenly your version feels small. You scroll X for 40 minutes. The original idea — which was fine — is now stained.
The move: Reality Check skill, then back to the work. The filter question: "if a VC-backed startup with 20 engineers is already doing this, why would I win?" If the answer is real, ship. If not, redirect.
A project is 85% done. The last 15% is naming things, writing docs, fixing the broken button on the contact page. The brain prefers to start a new shiny thing. The unfinished one becomes evidence against you.
The move: Ship-or-drift session. "Claude, what's the smallest thing I could do in 30 minutes that moves this from 85 → 90? No new features. Just close gaps."
Fiverr Gig #1 has been "30 minutes from live" for 35 days. Apr 25 and May 2 reviews both said the same thing: this isn't a time problem. There's something specific you don't want to do — and you haven't said it out loud yet.
The move: No more nudges. Open Claude. Type: "I keep avoiding the Fiverr gig. Help me name the specific thing I'm dodging — not the work, the feeling." Then act on the answer.
/rename. Future-you can't read your mind across three terminals.