Start Development Work
Begin work on a feature, bug fix, or enhancement. Establishes scope through adaptive questions, then you work. No ceremony - just clarity.
Part of the ritual: /pb-start → code → /pb-review → decide → /pb-commit
Mindset: Apply /pb-preamble thinking (challenge assumptions) and /pb-design-rules thinking (verify clarity, simplicity, robustness). This command ensures you know what success looks like before writing code.
Resource Hint: sonnet - Scope detection and branch setup
Voice: Conversational. System asks clarifying questions naturally, like a peer reviewing your plan. See /docs/voice.md for how commands communicate.
Tool-agnostic: This command works with any development tool or agentic assistant. Claude Code users invoke as /pb-start. Using another tool? Read this file as Markdown and work through the phases with your tool. See /docs/using-with-other-tools.md for adaptation examples.
When to Use
- Starting any new work (feature, fix, refactor)
- Need to clarify scope before coding
- Picking up work after a break (pair with
/pb-resume)
The Quick Start: 5 Minutes
/pb-start "feature name"
↓ System asks 3-4 adaptive questions
↓ You answer (1-2 min)
↓ Branch created, scope detected
↓ You code
What the conversation looks like:
The system asks clarifying questions naturally, like a peer reviewing your approach before you dive in. Adapt to what you describe:
-
What are you building? (outcome, not solution)
- You: “Users can reset passwords via email”
- System uses this to understand scope
-
How complex? (files and LOC estimate)
- You: “~200 LOC, 3 files, touches auth + email”
- System detects: small/medium/large
-
Scope mode? (expanding, holding, or reducing)
- Expanding: New capability - building something that doesn’t exist yet
- Holding: Hardening - bulletproofing, fixing, improving what exists
- Reducing: Surgical minimalism - removing, simplifying, cutting scope
- System adjusts review expectations: expanding gets architecture review, holding gets correctness review, reducing gets regression review
-
Critical path? (production, security, payment, or nice-to-have)
- You: “Payment processing, yes”
- System prepares review depth accordingly
-
How will you know it’s done? (verifiable success criteria)
- You: “Tests pass, can reset password end-to-end in staging”
- System records criteria;
/pb-reviewverifies against them
-
Any blockers?
- You: “Need staging DB access” or “None”
- System pauses if blockers exist, otherwise proceeds
After You Answer
System detects complexity level, criticality, and affected domains. Creates a feature branch with conventional naming, saves your scope and success criteria for /pb-review later, then gets out of your way. You code. No more decisions, no ceremony. System watches in the background, tracking change complexity as you work.
See /pb-llm-guidelines §4 for the goal-driven execution pattern: strong success criteria let you loop independently; weak criteria require constant clarification.
The Ritual is Simple
This command is part of a 3-command ritual:
/pb-start [what you're building]
↓ Answer 3-4 questions
↓ Branch created, scope recorded
[You code here-no interruptions]
/pb-review
↓ Detects review depth from your change
↓ Consults personas automatically
↓ Clean? Auto-commits. Issues? Preferences decide.
↓ Ambiguous? Asks you, then commits.
/pb-commit
↓ Usually automatic (triggered by /pb-review)
↓ Use explicitly if you want manual control
Total cognitive load: 3 commands. That’s a habit.
Pro Tips
Before you start:
- Read the outcome question carefully. “What are you building?” means outcome, not solution
- Be honest about complexity. Small estimate = lean review. Large = deep review.
- If blockers exist, resolve them now, don’t start coding with unknowns
After branch is created:
- Just code. Don’t think about the ritual yet.
- System is watching (tracking your changes)
- When done, run
/pb-review
Branch Naming
System auto-creates branch with conventional naming:
feature/short-descriptionfor new featuresfix/issue-descriptionfor bug fixesrefactor/what-changedfor refactoring
You don’t need to think about this.
Migration from Old Workflow
If you’ve used the playbook before, here’s what changed:
| Old | New |
|---|---|
/pb-start (long ceremony) | /pb-start (3-4 questions, 2 min) |
/pb-cycle (self-review) | /pb-review (auto-detects depth) |
/pb-review-code (peer review) | Built into /pb-review |
/pb-security, /pb-performance | Consulted automatically by /pb-review |
| Manual persona selection | Automatic (system decides who to consult) |
No more commands to remember: just /pb-start, /pb-review, /pb-commit.
Related Commands
/pb-review- Quality gate (the second part of the ritual)/pb-commit- Make the commit (the third part)/pb-pause- Pause work, save context/pb-resume- Get back into context/pb-plan- Plan architecture before starting (optional, for complex work)
One ritual. Three commands. Automagic depth detection. Quality by default.