A structured workflow for Claude Code that takes a plan and research-first approach to implementing features. This workflow provides skills, commands, and hooks to guide implementation from initial planning through to PR creation.
The Implementation Workflow is a comprehensive set of Claude Code skills and commands that help you:
- Plan systematically - Create detailed implementation plans with research, code snippets, and clear phases
- Execute methodically - Follow plans with incremental progress tracking and user confirmation
- Learn continuously - Capture corrections and learnings for future reference
- Integrate seamlessly - Work with GitHub issues, PRs, and git workflows
- iw-planner - Create detailed implementation plans through interactive research process
- iw-executor - Execute implementation plans with phase-based commits and tracking
- iw-workflow - Workflow guidance and documentation structure explanation
- iw-learnings - Search past learnings and corrections from previous work
- iw-init - Initialize .docs directory structure for plans and knowledge
- iw-github-issue-reader - Load comprehensive GitHub issue information
- iw-github-pr-creator - Create GitHub pull requests with plan summaries
- iw-git-workflow - Manage git operations with worktrees and phase commits
- iw-research-planner - Define research scope and create structured research plans
- iw-research-executor - Gather information and automatically generate comprehensive reports
- iw-research-synthesizer - Generate comprehensive reports (auto-invoked by executor)
- go-dev-guidelines - Go development patterns and TDD workflow (language-specific)
- skill-creator - Guide for creating new skills
- /iw-plan - Create detailed implementation plan
- /iw-implement - Execute implementation plan
- /iw-help - Show workflow guidance and available commands
- /iw-install - Manage workflow installation (install, update, uninstall)
- /iw-research-plan - Create new research plan with questions and scope
- /iw-research-execute - Gather information and generate comprehensive report
- load_workflow.py - Cross-platform session start hook showing documentation structure
- list_skills.py - Cross-platform hook to list available skills at session start
- check_workflow_version.py - Cross-platform hook to check for workflow updates
- load_workflow.sh - Deprecated bash version (kept for backward compatibility)
- list_skills.sh - Deprecated bash version (kept for backward compatibility)
- check_workflow_version.sh - Deprecated bash version (kept for backward compatibility)
- Python scripts for plan management, task tracking, and GitHub integration
- Template files for plan structure (plan, tasks, context, research)
- Reference documentation for Go development guidelines
The Implementation Workflow supports Windows, macOS, and Linux.
Recommended: Python Bootstrap (All Platforms)
# Download bootstrap script
curl -sSL https://raw.githubusercontent.com/jumppad-labs/iw/main/bootstrap.py -o bootstrap.py
# Run bootstrap
python3 bootstrap.pyAlternative: Bash Bootstrap (macOS/Linux Only)
# For macOS/Linux users who prefer bash
curl -sSL https://raw.githubusercontent.com/jumppad-labs/iw/main/bootstrap.sh | bashThe bootstrap will:
- Prompt you to choose project-level or user-level installation
- Download and set up the
iw-installskill - Provide instructions for completing installation
After bootstrap completes, restart Claude Code and run:
/iw-install
Windows:
- Requires Python 3.7+ and Git for Windows
- Use
python3 bootstrap.py(bootstrap.sh won't work without WSL) - PowerShell, Command Prompt, and Git Bash all supported
macOS/Linux:
- Both bootstrap.py and bootstrap.sh work
- Requires Python 3.7+ and git
- Python version recommended for consistency
You can install the workflow in two locations:
Install to .claude/ in your project root:
- Committed to version control
- Shared with team members
- Project-specific configuration
Install to ~/.claude/:
- Available in all projects
- Personal configuration
- Not committed to repos
For complex features or tasks that require research and structured planning:
/iw-plan 123
or
/iw-plan Add user authentication system
This will:
- Load GitHub issue context (if issue number provided)
- Search past learnings for relevant knowledge
- Launch parallel research agents to explore the codebase
- Guide you through context gathering and design decisions
- Create structured plan files in
.docs/issues/<number>/or.docs/adhoc/<name>/
After creating and reviewing a plan:
/iw-implement 123
or
/iw-implement .docs/issues/123/
This will:
- Load the plan files (plan.md, tasks.md, context.md, research.md)
- Create isolated worktree for implementation
- Execute tasks phase by phase
- Create commits after each phase
- Track progress in tasks.md
- Confirm with you at key milestones
/iw-help
Shows workflow guidance, documentation structure, and available commands.
For conducting systematic research on technical topics, academic papers, or code implementations. Now with Obsidian vault integration for seamless personal knowledge management.
The research workflow supports flexible storage locations:
- .docs/research/ - Standard project location (default)
- Obsidian vault - Auto-detected for personal knowledge base
- Custom paths - Any location you specify
Research happens in a temporary workspace, final report saved to your chosen location, and intermediate files are automatically cleaned up.
/iw-research-plan
This will:
- Detect Obsidian vault (if configured) and offer as workspace option
- Ask about your research topic and questions
- Help define 3-5 specific research questions
- Identify source types needed (papers, code repos, docs, articles)
- Define scope boundaries (included/excluded topics)
- Create research plan in chosen workspace location
/iw-research-execute
This will:
- Load your research plan
- Gather from multiple source types:
- Academic papers via WebSearch (Google Scholar, arXiv, etc.)
- Code repositories via GitHub exploration
- Technical documentation via official docs
- Blog posts and articles
- Organize findings by theme in
findings.md - Track all sources in
sources.md - Automatically generate comprehensive report:
- Analyze all gathered findings
- Generate executive summary (2-3 paragraphs)
- Organize key findings by theme
- Provide detailed analysis for each research question
- Include source references
- Add recommendations and next steps
- Create
research-report.mdin Obsidian-compatible markdown
Output: Comprehensive research report with:
- Executive summary
- Key findings organized by theme
- Detailed analysis with evidence
- Simple source citations
- Recommendations
If you have the obsidian-local-api skill configured:
-
Automatic vault detection:
- Skill detects your Obsidian vault automatically
- Offers vault root as workspace option
- Seamless integration with your knowledge base
-
Live preview during research:
- Work in vault root for live Obsidian preview
- See links and graph view while researching
- Leverage Obsidian features during work
-
Organized final storage:
- Choose final report location (e.g., Research/ folder)
- Intermediate files auto-cleaned
- Only final report remains in vault
Setup:
- Install Obsidian Local REST API plugin
- Configure via
obsidian-local-apiskill - Start research with
/iw-research-plan - Select Obsidian vault when prompted
Example workflow:
/iw-research-plan
# → Choose "Obsidian Vault Root" as workspace
# → Conduct research (files visible in Obsidian)
/iw-research-execute
# → Gather findings and sources
# → Final prompt: "Save report to Research/ folder"
# → Intermediate files auto-cleaned
# → Final report at: vault/Research/my-topic-report.mdThe typical workflow for a planned feature:
- Plan:
/iw-plan <issue-number>creates detailed implementation plan - Review: Review plan files, request changes if needed
- Implement:
/iw-implement <issue-number>executes the plan - Verify: Confirm after each phase, run tests
- Complete: Automatic PR creation and worktree cleanup
- Claude Code (latest version)
- Git (for version control operations)
- Git for Windows on Windows
- Standard git package on macOS/Linux
- Python 3.7+ (for helper scripts and installation)
- From python.org or Microsoft Store on Windows
- Pre-installed or via package manager on macOS/Linux
- GitHub CLI (gh) - For GitHub issue and PR operations
- jq - For JSON processing in scripts
- Go 1.21+ - If using go-dev-guidelines skill
- Windows 10/11 - Fully supported (PowerShell, Command Prompt, Git Bash)
- macOS - Fully supported
- Linux - Fully supported
If you used the bootstrap installation:
/iw-install --update
This fetches the latest version from GitHub and overwrites existing files.
Re-run the bootstrap command or manual installation steps to get the latest version.
/iw-install --uninstall
Remove the installed directories:
Project-level:
rm -rf .claude/skills/iw-*
rm -rf .claude/skills/go-dev-guidelines
rm -rf .claude/skills/skill-creator
rm -f .claude/commands/iw-*.md
rm -f .claude/hooks/load_workflow.sh
rm -f .claude/hooks/list_skills.shUser-level:
rm -rf ~/.claude/skills/iw-*
rm -rf ~/.claude/skills/go-dev-guidelines
rm -rf ~/.claude/skills/skill-creator
rm -f ~/.claude/commands/iw-*.md
rm -f ~/.claude/hooks/load_workflow.sh
rm -f ~/.claude/hooks/list_skills.shThe workflow creates a structured documentation approach:
.docs/
├── issues/ # Issue-based implementation plans
│ └── <number>/
│ ├── <number>-plan.md # Implementation plan
│ ├── <number>-tasks.md # Task checklist
│ ├── <number>-context.md # Quick reference
│ └── <number>-research.md # Research notes
├── adhoc/ # Ad-hoc plans not tied to issues
│ └── <name>/
│ └── (same structure)
└── knowledge/ # Institutional knowledge
├── learnings/ # Corrections and discoveries
├── architecture/ # Architecture documentation
└── gotchas/ # Known issues and workarounds
Create additional language-specific guideline skills following the pattern of go-dev-guidelines:
- Create
.claude/skills/<language>-guidelines/SKILL.md - Document coding standards, testing patterns, architecture patterns
- Reference from
iw-plannerandiw-executorskills
Use the skill-creator skill to create new custom skills.
- Verify files are in correct location (.claude/skills/)
- Check SKILL.md has proper YAML frontmatter
- Restart Claude Code session
- Ensure Python 3.7+ is installed
- Check scripts are in
scripts/subdirectory of skill - Verify script permissions (should be readable)
- Unix/macOS: Check bash hooks are executable:
chmod +x .claude/hooks/*.sh - Windows: Ensure Python hooks (.py files) are present in
.claude/hooks/ - All platforms: Verify hooks are in correct location
- Check hook configuration in settings
- Restart Claude Code to reload hooks
Unicode characters not displaying:
- Use Windows Terminal or PowerShell 7+ for better Unicode support
- Older Command Prompt may show fallback characters
- Scripts automatically handle encoding
Python scripts not executing:
- Verify Python 3.7+ is installed:
python3 --version - Ensure Python is in PATH
- Use
python3 script.pyinstead of./script.py
Contributions are welcome! Areas for contribution:
- Additional language-specific guidelines skills (Python, TypeScript, Rust, etc.)
- Enhanced GitHub integration features
- Additional helper scripts for common tasks
- Documentation improvements
Apache 2 - See LICENSE file for details
For issues, questions, or feedback:
- Open an issue on GitHub
- Check existing issues for similar problems
- Review workflow documentation in skills
Generated with Claude Code Implementation Workflow