Write the spec before touching the code
You want Claude to code the right thing on the first pass instead of iterating 5 times. Spec-First Development to set the exact frame before launching Claude, RPI Pattern to enforce the Research → Plan phase before implementation, RPI Commands to install the /rpi/research + /rpi/plan + /rpi/implement trio in .claude/commands/ with one clone.
Why
The agent codes fast. And it codes garbage if you've decided nothing. Classic scenario: you prompt "add a favorites system", the agent runs for 30 minutes, comes back with 12 files modified, a Supabase table, a custom hook, passing tests. Except the data model doesn't match what exists, the UX breaks onboarding, and you're deciding after the fact. You saved an hour, you'll lose six gluing it back together. This kit installs the opposite discipline: nothing gets coded until the spec is written. You split Research → Plan → Implement, each phase forks a dedicated sub-agent with fresh context.…
When
Framing phase before any non-trivial feature, for a dev/PM/tech-lead who's already eaten a vibe-code-rollback and won't do it again. ✅ You know it's your kit when You're starting a cross-cutting feature: touches schema, auth, multiple domains You rebase more than you ship: every PR rolls back 2-3 times You want to decide before the agent decides for you: data model, API contracts, UX flow ❌ Not…
Tools included
- Spec Kit
- Rpi Pattern
- Rpi Commands