Workflow commands for seamless end-to-end feature development with AI coding assistants. Works with Claude Code, OpenCode, and other AI development tools.
claudeflow v2.0.0 is a major release with breaking changes:
- Standalone package - No external dependencies (ClaudeKit/STM removed)
- Simplified installation - Just
npm installandclaudeflow setup - Tool-agnostic - Works with Claude Code, OpenCode, and other AI tools
- Lower requirements - Node.js 20+ (down from 22.14+)
- Streamlined workflow - Focus on workflow commands only
See CHANGELOG.md for full details.
Prerequisites: Node.js 20+ Check version:
node --version| Install: https://nodejs.org
Install claudeflow globally via your preferred package manager:
npm install -g @33strategies/claudeflowyarn global add @33strategies/claudeflowpnpm add -g @33strategies/claudeflowAfter installation, run the setup command:
claudeflow setupChoose your installation mode:
- 1) Global - Install to
~/.claude/(available in all projects) - 2) Project - Install to
./.claude/(this project only)
Quick decision:
- Solo developer or want commands everywhere? → Global mode
- Team project or want version control? → Project mode
- Both? → You can do both! They work together via configuration hierarchy.
The setup will:
- Check prerequisites (Node.js 20+, npm)
- Copy custom workflow commands to your chosen location
- Set up configuration files
claudeflow provides workflow commands for AI-assisted feature development:
- /brainstorm:start: Structured brainstorming with comprehensive documentation
- /brainstorm:clarify: Resolve open questions interactively
- /brainstorm:spec: Transform brainstorm into validated specification
- /feedback:add: Quick capture of feedback items after testing
- /feedback:resolve: Batch analyze and resolve pending feedback
- /spec:doc-update: Parallel documentation review based on specs
- /spec:create: Detects output path and creates specs in feature directories
- /spec:decompose: Incremental mode preserves completed work
- /spec:execute: Session resume capability with implementation tracking
- /spec:migrate: Migrates existing specs to feature-directory structure
- Complete end-to-end workflow from brainstorming to deployment
- Task tracking via
doc/specs/<slug>/03-tasks.md - Session continuity across implementation runs
- Interactive question resolution in specifications
Check that everything is set up correctly:
claudeflow doctorThis diagnostic command checks:
- Node.js version (need 20+)
- npm availability
- Claude Code CLI (optional - other AI tools supported)
- .claude/ directory structure
- Command files presence
- Ensure npm global bin is in your PATH
- Try:
npm list -g @33strategies/claudeflow - Reinstall:
npm install -g @33strategies/claudeflow
- Run:
claudeflow doctor - Verify files exist in
~/.claude/commands/(global) or./.claude/commands/(project) - Restart Claude Code
Run the diagnostic command for detailed information:
claudeflow doctorIf you previously used the bash script installation:
-
Uninstall old version:
# Global installation rm -rf ~/.claude # OR Project installation rm -rf .claude
-
Install via npm:
npm install -g @33strategies/claudeflow
-
Run setup:
# Global (if you used install.sh user) claudeflow setup --global # OR Project (if you used install.sh project) claudeflow setup --project
-
Verify installation:
claudeflow doctor
Note: The migration only affects the installation method, not your feature work or task data.
claudeflow is published with npm provenance attestations and SLSA Level 2 compliance. All releases are built and published via GitHub Actions with cryptographic verification.
For details on verifying package authenticity, supply chain security, and security best practices, see docs/SECURITY.md.
claudeflow checks for updates automatically once per week (every 7 days). This check runs in the background and doesn't block command execution.
Manual update check:
npm outdated -g @33strategies/claudeflowWhen updates are available, you'll see a notification:
╭───────────────────────────────────────────────────╮
│ │
│ Update available: 1.3.0 │
│ Current version: 1.2.0 │
│ Run: npm install -g @33strategies/claudeflow │
│ │
╰───────────────────────────────────────────────────╯
To update:
# npm
npm update -g @33strategies/claudeflow
# yarn
yarn global upgrade @33strategies/claudeflow
# pnpm
pnpm update -g @33strategies/claudeflowAll documents related to a feature are organized in a single directory:
doc/specs/
└── <feature-slug>/
├── 01-brainstorm.md # Brainstorming and research
├── 02-specification.md # Validated specification
├── 03-tasks.md # Task breakdown
├── 04-implementation.md # Implementation summary
└── 05-feedback.md # Post-implementation feedback log
Benefits:
- All related documents in one place
- Clear lifecycle progression (01 → 02 → 03 → 04)
- Easy to find and navigate
- Git-friendly tracking
Example:
doc/specs/add-user-auth-jwt/
├── 01-brainstorm.md # Created by /brainstorm:start
├── 02-specification.md # Created by /brainstorm:spec (after /brainstorm:clarify)
├── 03-tasks.md # Created by /spec:decompose
├── 04-implementation.md # Created by /spec:execute
└── 05-feedback.md # Created by /feedback:add (after testing)
claudeflow/ (@33strategies/claudeflow npm package)
├── package.json # npm package metadata
├── bin/
│ └── claudeflow.js # CLI entry point
├── lib/
│ ├── setup.js # Installation logic
│ └── doctor.js # Diagnostic command
├── scripts/
│ └── verify-files.js # Pre-publish file verification
│
├── .claude/ # Custom configuration (distributed in package)
│ ├── commands/ # Custom slash commands
│ │ ├── brainstorm/ # Brainstorming workflow commands
│ │ │ ├── start.md # Structured brainstorming workflow
│ │ │ ├── clarify.md # Resolve open questions
│ │ │ └── spec.md # Transform brainstorm to spec
│ │ └── spec/
│ │ ├── create.md # Enhanced spec creation
│ │ ├── decompose.md # Incremental task breakdown
│ │ ├── execute.md # Session-aware implementation
│ │ ├── feedback.md # Post-implementation feedback
│ │ ├── doc-update.md # Documentation review
│ │ └── migrate.md # Spec structure migration
│ ├── settings.json.example # Example configuration
│ └── README.md # Component documentation
│
├── .claude-plugin/ # Plugin metadata
│ └── plugin.json # Package metadata
│
├── templates/ # Configuration templates
│ ├── project-config/ # Project-level templates
│ │ ├── settings.json # Team settings
│ │ ├── settings.local.json.example
│ │ ├── CLAUDE.md # Project instructions
│ │ ├── CLAUDE.local.md.example
│ │ └── .gitignore.example
│ └── user-config/ # User-level templates
│ ├── settings.json # Personal global settings
│ └── CLAUDE.md # Personal preferences
│
├── docs/ # Documentation
│ ├── DESIGN_RATIONALE.md # Design validation & best practices
│ └── ...
│
├── .github/
│ └── workflows/
│ └── release.yml # CI/CD automation
├── .releaserc.json # Semantic-release config
├── .npmignore # Package exclusion rules
├── LICENSE # MIT License
├── CHANGELOG.md # Version history (auto-generated)
└── README.md # This file
Claude Code uses a 5-tier precedence system:
- Enterprise policies (highest) - IT-managed, cannot be overridden
- CLI arguments - Temporary session overrides
- Local settings (
.claude/settings.local.json) - Personal, gitignored - Project settings (
.claude/settings.json) - Team, committed - User settings (
~/.claude/settings.json) - Personal global defaults
Structured brainstorming workflow that enforces complete investigation for any code-change task (bug fix or feature). Produces comprehensive brainstorm documentation including:
- Intent & assumptions
- Pre-reading log
- Codebase mapping
- Root cause analysis (for bugs)
- Research findings
- Clarification questions
Usage: /brainstorm:start Fix chat UI auto-scroll bug when messages exceed viewport height
Resolve open questions in a brainstorm or specification document interactively. This command:
- Reads the document and detects unanswered questions
- Presents each question with context and options
- Records answers in strikethrough format (audit trail)
- Re-validates and loops until all questions resolved
- Shows progress: "Question 3 of 12"
Key Features:
- Supports multi-select for questions requiring multiple choices
- Detects external edits to prevent data loss
- Backward compatible (skips if no questions exist)
- Re-entrant (skips already-answered questions)
- Save-as-you-go for recoverability
Usage: /brainstorm:clarify doc/specs/<slug>/01-brainstorm.md
Example:
/brainstorm:clarify doc/specs/my-feature/01-brainstorm.md
# → System detects 5 open questions
# ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
# Question 1 of 5
# ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
# Package Manager Support
# → User selects: npm, yarn, pnpm (all three)
# → Document updated with strikethrough answer
# → Continues with questions 2-5
# → Final summary shows 5 questions resolvedAnswer Format (Audit Trail):
Questions are marked as resolved using strikethrough format:
Before:
1. **ClaudeKit Version Compatibility**
- Option A: Pin exact version
- Option B: Use caret rangeAfter:
1. ~~**ClaudeKit Version Compatibility**~~ (RESOLVED)
**Answer:** Use caret range (^1.0.0)
**Rationale:** Automatic updates, test compatibility in CI/CD
Original context preserved:
- Option A: Pin exact version
- Option B: Use caret rangeThis format provides a complete audit trail showing both the original question and the final decision.
Transform a brainstorm document into a validated, implementation-ready specification. This command:
- Reads and synthesizes the brainstorm document
- Interactively gathers decisions from the user
- Creates a detailed specification using
/spec:create - Validates it with
/spec:validate - Presents a summary with next steps
Usage: /brainstorm:spec doc/specs/<slug>/01-brainstorm.md
Note: Run /brainstorm:clarify first to resolve any open questions before creating the specification.
Quick capture of feedback items after manual testing. This command:
- Locates the specification (auto-selects most recent if not specified)
- Creates or loads
05-feedback.mdfile - Loops to capture multiple feedback items
- Saves each item immediately (save-as-you-go)
- Presents summary and next steps
Usage: /feedback:add or /feedback:add doc/specs/my-feature/02-specification.md
Example:
/feedback:add
# → Auto-selects most recent specification
# → "What feedback do you have?"
# → User enters: "Login button should be more prominent"
# → FB-1 added.
# → "What feedback do you have?" (loop continues)
# → User selects "Done"
# → Summary: 3 feedback item(s) added
# → Next: Run /feedback:resolve to analyze and resolveBatch analyze and resolve all pending feedback items. This command:
- Locates the feedback file (auto-selects most recent if not specified)
- Loads all pending items from
05-feedback.md - Analyzes all items in parallel using domain experts
- Presents each item with analysis for user decision (implement/defer/out-of-scope)
- Updates specification with approved changes
- Logs resolutions in
05-feedback.md
Integrates with incremental /spec:decompose and resume /spec:execute for seamless iteration.
Usage: /feedback:resolve or /feedback:resolve doc/specs/my-feature/05-feedback.md
Example:
/feedback:resolve
# → Auto-selects most recent feedback file
# → "3 pending feedback item(s)"
# → Analyzes all items in parallel
# → Presents FB-1 with analysis
# → User selects: "Implement"
# → Presents FB-2 with analysis
# → User selects: "Defer" (reason: "Not enough time")
# → Presents FB-3 with analysis
# → User selects: "Out of scope"
# → Updates specification with FB-1 changes
# → Summary: 1 implement, 1 defer, 1 out of scope
# → Next: Run /spec:decompose to create new tasksReview all documentation to identify what needs to be updated based on a new specification file. Launches parallel documentation expert agents to review each doc file for:
- Deprecated content
- Content requiring updates
- Missing content for new features
Usage: /spec:doc-update doc/specs/text-generator-spec.md
Creates specifications in feature directories at doc/specs/<slug>/02-specification.md.
Usage: /spec:create Add user authentication with JWT tokens
Breaks down specifications into tasks with incremental mode that preserves completed work. Creates task breakdown in doc/specs/<slug>/03-tasks.md.
Usage: /spec:decompose doc/specs/add-user-auth/02-specification.md
Incremental Mode: Automatically detects when spec changelog has been updated and creates only new tasks for changes.
Implements tasks with session resume capability. Reads implementation summary to skip completed tasks and continue from previous progress. Creates/updates doc/specs/<slug>/04-implementation.md.
Usage: /spec:execute doc/specs/add-user-auth/02-specification.md
Session Continuity: When you run /spec:execute multiple times on the same spec, it reads 04-implementation.md to understand what work has already been completed. This allows you to:
- Work on large features across multiple sessions without re-doing completed tasks
- Return to implementation after testing/feedback cycles
- Maintain a complete history of all sessions (Session 1, Session 2, etc.)
- Seamlessly integrate feedback workflow: implement → test →
/feedback:add→/feedback:resolve→/spec:decompose(incremental) →/spec:execute(resume)
Migrates existing specs from flat structure to feature-directory structure.
Usage: /spec:migrate
This repository implements a complete end-to-end workflow for feature development:
┌─────────────────────────────────────────────────────────────────┐
│ BRAINSTORM PHASE │
└─────────────────────────────────────────────────────────────────┘
│
▼
/brainstorm:start <task-brief>
(Custom Command)
│
Creates brainstorm doc
with research & analysis
│
▼
/brainstorm:clarify <brainstorm-doc>
(Resolve open questions)
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ SPECIFICATION PHASE │
└─────────────────────────────────────────────────────────────────┘
│
▼
/brainstorm:spec <brainstorm-doc>
(Custom, calls /spec:create & /spec:validate)
│
Creates validated specification
│
▼
/spec:decompose <spec-file>
│
Breaks spec into tasks
(creates 03-tasks.md)
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ IMPLEMENTATION PHASE │
└─────────────────────────────────────────────────────────────────┘
│
▼
/spec:execute <spec-file>
│
Implements tasks
(updates 03-tasks.md status)
│
▼
Manual Testing
│
Discover issues or
improvement opportunities
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ FEEDBACK PHASE │
└─────────────────────────────────────────────────────────────────┘
│
▼
/feedback:add (capture items)
│
Loop to add multiple items
(save-as-you-go)
│
▼
/feedback:resolve (batch analyze)
│
For each pending item:
• Parallel analysis
• User decision
│
┌──────────┴──────────┬────────────┐
│ │ │
Implement Now Defer Out of Scope
│ │ │
Update spec changelog Log for later Log only
│ │ │
└──────────┬──────────┴────────────┘
│
▼
/spec:decompose (incremental - if any "implement")
│
/spec:execute (resume)
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ COMPLETION PHASE │
└─────────────────────────────────────────────────────────────────┘
│
▼
/spec:doc-update <spec-file>
│
Reviews & updates documentation
│
▼
git commit & push
│
▼
DONE! 🎉
- Brainstorm → Comprehensive investigation and research
- Clarify → Resolve open questions interactively
- Specification → Validated, implementation-ready spec
- Decomposition → Tasks broken down in
03-tasks.md - Implementation → Iterative execution with task status tracking
- Feedback → Process post-implementation feedback with structured decisions (implement/defer/out-of-scope)
- Completion → Documentation updates and git workflow
# Step 1: Start with brainstorming
/brainstorm:start Add user authentication with JWT tokens
# → Creates: doc/specs/add-user-auth-jwt/01-brainstorm.md
# Step 2: Resolve any open questions
/brainstorm:clarify doc/specs/add-user-auth-jwt/01-brainstorm.md
# → Interactively resolves all open questions
# Step 3: Transform to validated specification
/brainstorm:spec doc/specs/add-user-auth-jwt/01-brainstorm.md
# → Creates: doc/specs/add-user-auth-jwt/02-specification.md (validated & complete)
# Step 4: Break down into tasks
/spec:decompose doc/specs/add-user-auth-jwt/02-specification.md
# → Creates: doc/specs/add-user-auth-jwt/03-tasks.md
# Step 5: Start implementation
/spec:execute doc/specs/add-user-auth-jwt/02-specification.md
# → Implements tasks, updates status in 03-tasks.md
# Step 6: Continue implementing (loop back to step 5 if needed)
/spec:execute doc/specs/add-user-auth-jwt/02-specification.md
# → Resume mode: skips completed work
# Step 7: Manual testing (discover feedback items)
# Test the implemented feature manually
# Step 8: Capture feedback items
/feedback:add doc/specs/add-user-auth-jwt/02-specification.md
# → Loop to capture multiple feedback items
# → Each item saved immediately
# Step 9: Resolve feedback (batch analyze and decide)
/feedback:resolve doc/specs/add-user-auth-jwt/05-feedback.md
# → Analyzes all pending items in parallel
# → Interactive decisions: implement/defer/out-of-scope
# → Updates spec with approved changes
# Step 10: If any "Implement" decisions, run incremental decompose
/spec:decompose doc/specs/add-user-auth-jwt/02-specification.md
# → Incremental mode: preserves completed tasks, creates only new ones
# Step 11: Resume implementation for new tasks
/spec:execute doc/specs/add-user-auth-jwt/02-specification.md
# Step 12: Update documentation and commit
/spec:doc-update doc/specs/add-user-auth-jwt/02-specification.md
# git add . && git commit -m "feat: add user authentication"
# git push
# All documents for this feature are now in: doc/specs/add-user-auth-jwt/If you already know what you need:
# Start directly with spec creation
/spec:create Add user authentication with JWT tokens
# Then follow steps 4-12 aboveIf you have specs in the old flat structure:
# Migrate all existing specs to new structure
/spec:migrate
# This will:
# - Move specs/*.md to doc/specs/<slug>/02-specification.md
# - Move specs/*-tasks.md to doc/specs/<slug>/03-tasks.md
# - Move docs/ideation/*.md to doc/specs/<slug>/01-brainstorm.md
# - Generate migration report# After implementing a feature via a spec
/spec:doc-update doc/specs/add-user-auth-jwt/02-specification.md
# This will:
# - Review all docs for outdated content
# - Identify missing documentation for new features
# - Provide prioritized recommendationsProject Settings (.claude/settings.json - committed):
{
"permissions": {
"allow": ["Read", "Edit", "Bash(git:*)"],
"deny": [".env", "**/*.key", "secrets/"]
}
}Personal Overrides (.claude/settings.local.json - gitignored):
- Use for personal preferences that shouldn't be committed
- Never commit secrets to
.claude/settings.json - Use environment variable references:
"API_KEY": "${API_KEY}" - Keep
.envfiles gitignored
- Commit
.claude/settings.jsonandCLAUDE.mdto git - Gitignore
.claude/settings.local.jsonandCLAUDE.local.md
- Keep CLAUDE.md focused and concise
- Break large tasks into smaller steps
- Installation Guide - When to use global vs project installation
- Setup Guide - Detailed setup instructions
- Design Rationale - Design validation and best practices
- Feedback Workflow Guide - Complete guide to post-implementation feedback
- GitHub Issues - Report problems
Contributions welcome! Please:
- Follow existing patterns and structure
- Test changes thoroughly
- Update documentation
- Submit PRs with clear descriptions
MIT License - feel free to use and modify for your projects.
- Anthropic - Claude Code CLI and AI development
- Community contributors and examples