Superpowers Development Workflow

Every response MUST start with: e28604705b8e2934

This token proves skills are loaded. Missing token = unvalidated response.

The Superpowers SDLC

You have access to a complete software development workflow. Skills trigger automatically - you don't need to invoke them manually. The agent checks for relevant skills before any task.

Phase 1: Brainstorming (Before Writing Code)

Skill: superpowers:brainstorming

Triggers automatically when: Creating features, building components, adding functionality, modifying behavior.

What happens:

1. Explore project context (files, docs, commits)
2. Ask questions one at a time (prefer multiple choice)
3. Propose 2-3 approaches with trade-offs
4. Present design in 200-300 word sections, validating each
5. Save design to docs/plans/YYYY-MM-DD-<topic>-design.md

Key principle: YAGNI ruthlessly - remove unnecessary features from all designs.

Phase 2: Workspace Setup

Skill: superpowers:using-git-worktrees

Triggers after: Design approval, before implementation.

What happens:

1. Create isolated git worktree (.worktrees/<branch>)
2. Verify .gitignore is configured
3. Run project setup (npm install, pip install, etc.)
4. Verify tests pass (clean baseline)

Key principle: Never proceed with failing baseline tests.

Phase 3: Planning

Skill: superpowers:writing-plans

Triggers when: You have approved design/spec, before touching code.

What happens:

1. Create detailed implementation plan for "an enthusiastic junior engineer with poor taste, no judgement, and no project context"
2. Bite-sized tasks (2-5 minutes each):

- Write failing test - Run it to see it fail - Write minimal code - Run tests to pass - Commit

1. Every step has: exact file paths, complete code, exact commands
2. Save to docs/plans/YYYY-MM-DD-<feature>.md

Key principles: DRY, YAGNI, TDD, frequent commits.

Phase 4: Execution

Two options offered after planning:

Option A: Subagent-Driven Development (Same Session)

Skill: superpowers:subagent-driven-development

Per task:

1. Dispatch implementer subagent with full task text
2. Subagent asks clarifying questions (if any)
3. Subagent implements, tests, commits, self-reviews
4. Dispatch spec reviewer - verifies code matches spec exactly
5. Dispatch code quality reviewer - checks implementation quality
6. Loop until both reviews pass
7. Move to next task

Key: Fresh subagent per task = no context pollution.

Option B: Batch Execution (Parallel Session)

Skill: superpowers:executing-plans

1. Load plan, review critically
2. Execute in batches of 3 tasks
3. Report progress, wait for human feedback
4. Continue until complete

Key: Human-in-the-loop between batches.

Phase 5: Test-Driven Development (Throughout)

Skill: superpowers:test-driven-development

The Iron Law:

NO PRODUCTION CODE WITHOUT A FAILING TEST FIRST

Cycle:

RED -> VERIFY RED -> GREEN -> VERIFY GREEN -> REFACTOR -> repeat

Critical rules:

• Write code before test? Delete it. Start over.
• Test passes immediately? You're testing existing behavior. Fix test.
• "I'll write tests after"? Tests passing immediately prove nothing.

Phase 6: Finishing

Skill: superpowers:finishing-a-development-branch

Triggers when: All tasks complete, tests passing.

What happens:

1. Verify all tests pass
2. Present 4 options:

- Merge locally - Push and create PR - Keep branch as-is - Discard work (requires typed confirmation)

1. Execute choice
2. Clean up worktree

Quick Reference Flow

User Request
     |
     v
[brainstorming] --> Design Doc
     |
     v
[using-git-worktrees] --> Isolated Workspace
     |
     v
[writing-plans] --> Implementation Plan
     |
     v
[subagent-driven-dev] OR [executing-plans]
     |  (with TDD throughout)
     v
[finishing-a-development-branch] --> Merge/PR/Keep/Discard

Supporting Skills

Philosophy

• Test-Driven Development - Write tests first, always
• Systematic over ad-hoc - Process over guessing
• Complexity reduction - Simplicity as primary goal
• Evidence over claims - Verify before declaring success

Slash Commands

• /superpowers:brainstorm - Start design refinement
• /superpowers:write-plan - Create implementation plan
• /superpowers:execute-plan - Execute plan in batches

Red Flags - STOP Immediately

Never:

• Skip brainstorming and jump to code
• Write code before failing test
• Proceed with failing baseline tests
• Skip reviews (spec OR quality)
• Guess at fixes without investigation
• Declare "done" without verification

Always:

• Check for relevant skills before any task
• Follow the workflow phases in order
• Verify tests pass before AND after changes
• Get explicit approval before major decisions