Start-Work is a complete Claude Code planning solution for solo developers and small teams. Initialize work items, create detailed phase plans, and execute with integrated workflows.
Not sure how to start with Claude Code? Just Start-Work!
This package includes two complementary Claude Code skills:
start-work- Initialize structured work items (like JIRA issues) with auto-numberingcreate-plans- Create hierarchical, executable project plans optimized for Claude
Plus three slash commands: /start-work, /create-plan, and /run-plan
npx @prillcode/start-workThat's it! The installer will:
- Install both skills to
~/.claude/skills/ - Install slash commands to
~/.claude/commands/ - Back up any existing installations
- Verify everything is working
Creates structured planning directories for features, bugs, refactors, and enhancements.
Features:
- Auto-numbered identifiers (
ID-01,ID-02) or custom (mdo-123,gh-456) - Work-level BRIEF.md and ROADMAP.md templates
- Integration with external tracking systems (JIRA, GitHub, etc.)
- Duplicate detection with auto-suffixing
- Optional project-level documentation
Usage:
# Interactive mode
/start-work
# With arguments
/start-work mdo-123 api-refactor # Custom ID
/start-work "" parallels-ui-updates # Auto-number
/start-work gh-456 performance-optimization # GitHub issueCreates:
.planning/
└── mdo-123-api-refactor/
├── BRIEF.md # Work scope and vision
├── ROADMAP.md # Phase breakdown
└── phases/ # Execution plans go here
Create executable plans that Claude can run, not enterprise documentation that sits unused.
Philosophy:
- Plans are written AS prompts (PLAN.md IS the execution prompt)
- Optimized for solo developer + Claude (no teams, no stakeholders)
- Scope control keeps plans within ~50% context usage
Features:
- BRIEF.md → Human vision (what and why)
- ROADMAP.md → Phase structure
- PLAN.md → Claude-executable prompts
- SUMMARY.md → Execution outcomes
- Research workflow for unknowns
- Checkpoint management for human input
Usage:
# Interactive planning
/create-plan
# The skill guides you through:
# - Scanning existing structure
# - Creating briefs and roadmaps
# - Generating executable plansExecute PLAN.md files with intelligent segmentation for optimal quality.
Strategies:
- Fully Autonomous - No checkpoints, spawn single subagent
- Segmented Execution - Verify-only checkpoints, parallel segments
- Decision-Dependent - Action checkpoints, sequential in main context
Usage:
/run-plan .planning/mdo-123-api-refactor/phases/01-01-PLAN.md/start-work mdo-123 "API Refactor"Creates .planning/mdo-123-api-refactor/ with BRIEF.md and ROADMAP.md
Edit the generated files:
- Update BRIEF.md with specific scope and success criteria
- Refine ROADMAP.md with phase breakdown
cd .planning/mdo-123-api-refactor
/create-planGenerates executable PLAN.md files in phases/ directory
/run-plan .planning/mdo-123-api-refactor/phases/01-01-PLAN.mdClaude executes the plan, creates SUMMARY.md, and commits changes
.planning/
├── BRIEF.md # Project-level (optional)
├── ROADMAP.md # Project-level (optional)
├── ID-01-feature/ # Auto-numbered work
│ ├── BRIEF.md
│ ├── ROADMAP.md
│ └── phases/
│ ├── 01-01-PLAN.md
│ └── 01-01-SUMMARY.md
├── mdo-123-api-refactor/ # Custom identifier
│ ├── BRIEF.md
│ ├── ROADMAP.md
│ └── phases/
│ ├── 01-01-PLAN.md
│ ├── 01-01-SUMMARY.md
│ ├── 01-02-PLAN.md
│ └── 01-02-SUMMARY.md
└── ID-02-bug-fix/ # Another auto-numbered work
├── BRIEF.md
├── ROADMAP.md
└── phases/
| User Input | Work Name | Generated Directory |
|---|---|---|
| (blank) | parallels-ui-updates |
ID-01-parallels-ui-updates/ |
| (blank) | fix-auth-bug |
ID-02-fix-auth-bug/ |
mdo-123 |
api-refactor |
mdo-123-api-refactor/ |
gh-456 |
performance-optimization |
gh-456-performance-optimization/ |
JIRA-789 |
database-migration |
JIRA-789-database-migration/ |
✅ Auto-numbering - Scans .planning/ for ID-[nn] patterns, generates next number
✅ Custom identifiers - Support for JIRA, GitHub, or any tracking system
✅ Duplicate handling - Auto-appends -02, -03 if directory exists
✅ Templates - Work-level and project-level BRIEF/ROADMAP templates
✅ Context detection - Different commit behavior for user vs subagent invocation
✅ Plans are prompts - PLAN.md IS the execution prompt, not documentation ✅ Scope control - Plans complete within ~50% context for consistent quality ✅ Research workflow - Optional RESEARCH.md for unknowns ✅ Checkpoints - Human verification, decision, and action checkpoints ✅ Verification - All tasks include verification criteria ✅ Summaries - Structured SUMMARY.md documents outcomes
Skills:
~/.claude/skills/start-work/- Work item initialization skill~/.claude/skills/create-plans/- Hierarchical planning skill
Commands:
~/.claude/commands/start-work.md- Invokes start-work skill~/.claude/commands/create-plan.md- Invokes create-plans skill~/.claude/commands/run-plan.md- Executes PLAN.md files (standalone)
Templates:
- Work-level BRIEF.md and ROADMAP.md
- Project-level BRIEF.md and ROADMAP.md
- Planning references and workflows
# Check skills
ls ~/.claude/skills/start-work/SKILL.md
ls ~/.claude/skills/create-plans/SKILL.md
# Check commands
ls ~/.claude/commands/start-work.md
ls ~/.claude/commands/create-plan.md
ls ~/.claude/commands/run-plan.mdSimply run the installer again:
npx @prillcode/start-workExisting installations are automatically backed up before updating.
feat: implement new feature
fix: resolve authentication bug
refactor: improve code structure
docs: update documentationfeat(mdo-123-01): implement API endpoint
fix(ID-02-01): resolve authentication bug
refactor(gh-456-02): optimize database queries- Claude Code CLI installed with compatible Node version
- Git (recommended for version control)
# Verify installation
ls ~/.claude/skills/start-work/
ls ~/.claude/skills/create-plans/
# If missing, reinstall
npx @prillcode/start-work# Check if commands exist
ls ~/.claude/commands/*.md
# Make sure you're using forward slashes
/start-work (not \start-work)The directory is created automatically when you run /start-work. Check file permissions in your working directory.
For solo developers and small teams, not enterprise planning processes.
- Plans are prompts, not documentation
- Scope control prevents context bloat
- Human checkpoints for critical decisions
- Verification-first approach ensures quality
- Executable from day one
Issues and pull requests welcome at github.com/prillcode/start-work
MIT
- create-plans skill originally from glittercowboy/taches-cc-resources
- Maintained and enhanced by prillcode
- GitHub: github.com/prillcode/start-work
- Issues: github.com/prillcode/start-work/issues
- NPM: @prillcode/start-work
Made with ❤️ for Claude Code developers