claude-dev-cli 0.20.0

A powerful CLI tool for developers using Claude AI with multi-API routing, test generation, code review, usage tracking, and project management automation

Claude Dev CLI

⚠️ NOTICE: This project is being renamed to devflow in v0.20.0!
The cdc command will continue to work. Migration Guide | Rename Details

A powerful command-line tool for developers using Claude AI with multi-API routing, test generation, code review, comprehensive usage tracking, and AI-powered project automation.

Features

🧠 Intelligent Context Gathering (v0.18.0+) ✨ NEW

AI learns your codebase patterns before generating code:

# Automatically gathers context (language, framework, conventions, similar code)
cdc ticket execute TASK-123 --commit --notify

# Skip context gathering for faster execution
cdc ticket execute TASK-456 --no-context

What context is gathered:

• 🗂️ Language & Framework (Python/Django, JS/React, Go, etc.)
• 📦 Dependencies (requirements.txt, package.json, go.mod)
• 📁 Directory Structure (models/, views/, controllers/, tests/)
• 📝 Naming Conventions (snake_case, camelCase, PascalCase)
• 🔍 Similar Existing Code (based on ticket keywords)
• ⚙️ Common Imports & Patterns

Result: AI generates code that follows YOUR project's patterns, not generic templates!

See CONTEXT_INTELLIGENCE.md for full details.

🤖 AI-Powered Project Automation (v0.17.0+)

Transform tickets into working code automatically:

# Execute a ticket end-to-end: fetch → analyze → generate code → tests → commit
cdc ticket execute TASK-123 --commit --notify

What happens:

1. 📥 Fetches ticket from backend (repo-tickets, Jira, Linear, GitHub Issues)
2. 🧠 AI analyzes requirements and acceptance criteria
3. 💻 Generates production-ready implementation code
4. ✅ Generates comprehensive tests
5. 📝 Commits to git with conventional format + co-author
6. 🎫 Updates ticket status to "completed"
7. 📊 Logs all progress to .cdc-logs/progress.md
8. 🔔 Sends push notifications (FREE via ntfy.sh)

Ticket Management:

# Create tickets
cdc tickets create "Add user authentication" --priority high --type feature

# List tickets
cdc tickets list --status open

# Multi-backend support
cdc ticket execute TASK-456 --backend repo-tickets

AI-Powered Bug Triage:

# Report bugs with automatic severity/category classification
cdc bug report \
  --title "App crashes on startup" \
  --description "Application fails to launch" \
  --expected "App should start normally" \
  --actual "App crashes immediately" \
  --steps "Open app" --steps "Wait 2 seconds" \
  --environment production

# AI automatically assigns:
# ✅ Severity: CRITICAL (crash detected)
# ✅ Category: CRASH
# ✅ Priority: critical
# ✅ Sends urgent notification
# ✅ Detects duplicates

Progress Tracking:

# Initialize logging
cdc log init "My Project"

# Add entries
cdc log entry "Completed auth module" --ticket TASK-123 --level success

# Generate reports
cdc log report  # Beautiful markdown summary with stats

Features:

• 🎫 Ticket Backends: repo-tickets, Jira, Linear, GitHub Issues, Markdown (fallback)
• 🤖 Full AI Automation: Requirements → Code → Tests → Commit → Deploy
• 🐛 Smart Bug Triage: AI classifies severity, category, priority
• 📊 Progress Logging: Markdown-based tracking with artifact linking
• 🔔 Notifications: FREE push notifications via ntfy.sh (no signup!)
• 🔄 VCS Integration: Git commits with co-author attribution
• 📝 Structured Data: BDD scenarios, acceptance criteria, stack traces
• 🔍 Duplicate Detection: Find similar bugs automatically
• ⚡ Cost Optimization: Use FREE local AI (Ollama) for code generation

Install with project management:

pip install 'claude-dev-cli[all-pm]'  # All project management features
pip install 'claude-dev-cli[project-mgmt]'  # Ticket management only

🌐 Multi-Provider AI Support (v0.14.0+)

• Anthropic (Claude): GPT-4 class models with 200k context
  • Haiku, Sonnet, Opus - full model family
  • Industry-leading context window
• OpenAI (GPT): ChatGPT and GPT-4 models (v0.15.0+)
  • GPT-3.5 Turbo, GPT-4, GPT-4 Turbo
  • Azure OpenAI support
  • Install: pip install 'claude-dev-cli[openai]'
• Ollama (Local): Zero-cost local inference (v0.16.0+)
  • Mistral, Mixtral, Code Llama, DeepSeek Coder
  • 100% free, private, offline
  • Install: pip install 'claude-dev-cli[ollama]'
• Provider Abstraction: Unified interface across all providers
• Cost Tracking: Per-provider usage and cost monitoring
• Graceful Fallback: Works with any combination of providers

🔑 Multi-API Key Management

• Secure Storage: API keys stored in system keyring (macOS Keychain, Linux Secret Service, Windows Credential Locker)
• Route tasks to different API keys/providers (personal, client, enterprise, local)
• Automatic API selection based on project configuration
• Automatic migration from plaintext to secure storage

🎯 Model Profiles (v0.10.0+)

• Named Profiles: Use friendly names instead of full model IDs (fast, smart, powerful)
• Custom Pricing: Define input/output costs per Mtok for accurate usage tracking
• API-Specific Profiles: Different models and pricing per API config
• Project Defaults: Set per-project model preferences
• Dynamic Resolution: Profile names automatically resolve to model IDs
• Built-in Profiles:
  • fast: Claude 3.5 Haiku ($0.80/$4.00 per Mtok)
  • smart: Claude Sonnet 4 ($3.00/$15.00 per Mtok) - default
  • powerful: Claude Opus 4 ($15.00/$75.00 per Mtok)

🚀 Code Generation (v0.12.0+)

• Generate Code from Specs: Create new code from descriptions, files, PDFs, or URLs
  • cdc generate code --description "REST API client" -o client.py
  • Multiple input sources: text, files (.md, .txt), PDFs, URLs
  • Auto-detects target language from file extension
  • Interactive refinement mode
• Add Features to Projects: Analyze existing code and generate implementation plans
  • cdc generate feature --description "Add authentication" src/
  • Multi-file analysis and modification
  • Preview mode to review changes before applying
  • Supports same input sources as code generation
• Multiple Input Sources:
  • --description TEXT: Inline specification
  • -f/--file PATH: Read from file
  • --pdf PATH: Extract from PDF
  • --url URL: Fetch from URL
• Optional Dependencies: Install with pip install 'claude-dev-cli[generation]'
  • Enables PDF and URL support
  • Graceful fallback if not installed

📁 Multi-File Support (v0.11.0+)

• Batch Processing: Review, refactor, test, or document multiple files at once
• Directory Support: Process all code files in a directory with --max-files limit
• Auto-Detection: Commands auto-detect git changes when no files specified
  • cdc review → reviews staged files, falls back to modified files, then current directory
• Git Integration: New cdc git review command for reviewing changes
  • --staged: Review only staged changes
  • --branch <range>: Review branch changes (e.g., main..HEAD)
• Multi-Language: Supports 25+ file extensions (Python, JS/TS, Go, Rust, Java, C++, etc.)
• Smart Display: Shows file list preview (first 5-10 files, then "... and N more")
• Commands with multi-file support:
  • review, refactor, generate tests, generate docs

🧪 Developer Tools

• Test Generation: Automatic pytest test generation for Python code
• Code Review: Comprehensive code reviews with security, performance, and best practice checks
• Debug Assistant: Analyze errors and get fixes with explanations
• Documentation: Generate docstrings and README documentation
• Refactoring: Get suggestions for code improvements
• Git Integration: Generate conventional commit messages from diffs

📊 Usage Tracking

• Track token usage per API key
• Cost estimation based on current pricing
• Filter by date range and API config
• Detailed breakdowns by model, date, and API

💬 Interaction Modes

• Single-shot: Quick questions with piping support
• Interactive: Chat mode with conversation history
• Streaming: Real-time responses

📝 Custom Templates

• Built-in Templates: 8 pre-built templates for common tasks (code review, testing, debugging, etc.)
• User Templates: Create and manage your own reusable prompt templates
• Variable Substitution: Use {{variable}} placeholders for dynamic content
• Categories: Organize templates by category (review, testing, debugging, etc.)

🧠 Context Intelligence (v0.8.0+)

• Auto-Context: --auto-context flag on 7 commands for intelligent context gathering
  • ask, review, debug, refactor (v0.8.0)
  • git commit, generate tests, generate docs (v0.8.1)
• Git Integration: Automatically include branch, commits, modified files
• Dependency Analysis: Parse imports and include related files
• Multi-Language Error Parsing (v0.8.2): Python, JavaScript/TypeScript, Go, Rust, Java
• Context Summary (v0.8.2): Preview context before API calls with cdc context summary
• Smart Truncation: Prevent token limits with configurable file size limits
• Project Memory: Remember preferences per project
• Global Config: Set context defaults in ~/.claude-dev-cli/config.json

🔄 Workflows (v0.7.0+, Enhanced v0.16.1)

• Multi-Step Automation: Chain AI operations and shell commands with YAML
• AI Decision Making: Use ask command to make intelligent workflow decisions (v0.16.1)
• Variable Interpolation: Pass data between steps with {{variable}} syntax
• Conditional Execution: Skip steps based on conditions with if clauses
• Approval Gates: Human-in-the-loop with approval_required: true
• Model Selection: Use different AI models per step for cost optimization (v0.16.1)
• Error Handling: Continue on failure with continue_on_error: true
• Three Step Types: command (cdc), shell (bash/zsh), set (variables)
• Cost Optimization: Mix free local AI with paid cloud AI strategically

🎒 TOON Format Support (Optional)

• Token Reduction: 30-60% fewer tokens than JSON
• Cost Savings: Reduce API costs significantly
• Format Conversion: JSON ↔ TOON with CLI tools
• Auto-fallback: Works without TOON installed

Installation

Via Homebrew (macOS/Linux)

# Add the tap
brew tap thinmanj/tap

# Install
brew install claude-dev-cli

# Or in one command
brew install thinmanj/tap/claude-dev-cli

Via pip

Basic Installation

Core dependencies only (includes unidiff for diff parsing):

pip install claude-dev-cli

Core dependencies:

• anthropic>=0.18.0 - Claude API client
• click>=8.1.0 - CLI framework
• rich>=13.0.0 - Terminal formatting
• pydantic>=2.0.0 - Data validation
• keyring>=24.0.0 - Secure credential storage
• cryptography>=41.0.0 - Encryption for secure storage
• pyyaml>=6.0.0 - YAML configuration support
• unidiff>=0.7.0 - Production-grade diff parsing

With Optional Features

Multi-Provider Support:

# OpenAI (GPT) support
pip install 'claude-dev-cli[openai]'

# Ollama (local/free) support
pip install 'claude-dev-cli[ollama]'

# All providers
pip install 'claude-dev-cli[all-providers]'

Code Generation Support (PDF & URL input):

pip install 'claude-dev-cli[generation]'

Adds:

• pypdf>=3.0.0 - PDF text extraction
• requests>=2.28.0 - HTTP client for URL fetching
• beautifulsoup4>=4.0.0 - HTML parsing

TOON Format Support (30-60% token reduction):

pip install 'claude-dev-cli[toon]'

Adds:

• toon-format>=0.1.0 - TOON encoding/decoding

Project Management (v0.17.0+ - ticket automation, bug triage):

# Core project management (tickets, logging)
pip install 'claude-dev-cli[project-mgmt]'

# With notifications
pip install 'claude-dev-cli[notifications]'

# With VCS integration
pip install 'claude-dev-cli[vcs]'

# All project management features
pip install 'claude-dev-cli[all-pm]'

Adds:

• repo-tickets>=0.8.0 - Ticket management backend
• python-docx>=0.8.11 - Document processing
• requests>=2.28.0 - HTTP client for notifications
• python-telegram-bot>=20.0 - Telegram notifications (optional)
• GitPython>=3.1.0 - Git integration
• docker>=6.0.0 - Docker environment support

Syntax Highlighting (enhanced diff display):

pip install 'claude-dev-cli[plugins]'

Adds:

• pygments>=2.0.0 - Syntax highlighting for diffs

All Optional Features:

pip install 'claude-dev-cli[generation,toon,plugins,all-pm]'

Development Installation (for contributors):

pip install 'claude-dev-cli[dev]'

Includes all optional features plus:

• pytest>=7.0.0 - Testing framework
• black>=23.0.0 - Code formatting
• ruff>=0.1.0 - Linting
• mypy>=1.0.0 - Type checking

Via pipx (Recommended for CLI tools)

pipx provides isolated installations without affecting your system Python:

# Basic installation
pipx install claude-dev-cli

# With code generation support (PDF & URL)
pipx install 'claude-dev-cli[generation]'

# With TOON support (token reduction)
pipx install 'claude-dev-cli[toon]'

# With syntax highlighting
pipx install 'claude-dev-cli[plugins]'

# With all optional features
pipx install 'claude-dev-cli[generation,toon,plugins]'

Upgrade Existing Installation

# Upgrade via pip
pip install --upgrade claude-dev-cli

# Upgrade via pipx
pipx upgrade claude-dev-cli

# Upgrade via Homebrew
brew upgrade claude-dev-cli

Quick Start

1. Set Up API Keys

# Add your personal API key
export PERSONAL_ANTHROPIC_API_KEY="sk-ant-..."
cdc config add personal --default --description "My personal API key"
# 🔐 Stored securely in system keyring

# Add client's API key
export CLIENT_ANTHROPIC_API_KEY="sk-ant-..."
cdc config add client --description "Client's Enterprise API"

# List configured APIs (shows storage method)
cdc config list

# Manually migrate existing keys (automatic on first run)
cdc config migrate-keys

1.1 Setup Ollama (Local/Free Alternative) - v0.16.0+

Run AI models locally with zero cost and full privacy using Ollama:

# Install Ollama (one-time setup)
# macOS/Linux: Download from https://ollama.ai
# Or via Homebrew: brew install ollama

# Start Ollama server
ollama serve  # Run in background or separate terminal

# Install Ollama support
pip install 'claude-dev-cli[ollama]'

# Configure local provider (no API key needed!)
cdc config add ollama local --default
# ℹ️  No API key needed for local provider

# Pull models (one-time per model)
cdc ollama pull mistral      # Fast, general-purpose (7B)
cdc ollama pull codellama    # Code-specialized (7B-34B)
cdc ollama pull mixtral      # Powerful reasoning (8x7B)

# List available models
cdc ollama list
# ┌──────────────┬─────────────────┬─────────┬──────────┬─────────────────┐
# │ Model        │ Display Name    │ Context │ Cost     │ Capabilities    │
# │ mistral      │ Mistral 7B      │ 8,192   │ FREE     │ chat, code      │
# │ codellama    │ Code Llama      │ 16,384  │ FREE     │ code, chat      │
# │ mixtral      │ Mixtral 8x7B    │ 32,768  │ FREE     │ chat, analysis  │
# └──────────────┴─────────────────┴─────────┴──────────┴─────────────────┘

# Show model details
cdc ollama show mistral

Benefits of Local Models:

• ✅ Zero Cost: No API fees, unlimited usage
• ✅ Privacy: Data never leaves your machine
• ✅ Offline: Works without internet connection
• ✅ Fast: No API latency (after model loads)
• ✅ Control: Full control over models and data

Considerations:

• ⚠️ Requires decent hardware (8GB+ RAM, GPU recommended)
• ⚠️ Models need disk space (4-40GB per model)
• ⚠️ Quality may be lower than GPT-4/Claude Opus
• ⚠️ Initial model download can take time

Use Local Models:

# Use with any command - completely FREE!
cdc ask "explain async/await"
cdc ask -m fast-local "quick question"
cdc generate tests -m code-local mymodule.py
cdc review -m smart-local src/

# Use specific model directly
cdc ask -m mistral "your question"
cdc ask -m codellama "write a function to parse JSON"

# Remote Ollama server
cdc config add ollama remote --base-url http://server:11434
cdc ask -a remote "question"

Available Model Profiles:

• fast-local: mistral (8k context, fast inference)
• smart-local: mixtral (32k context, powerful)
• code-local: codellama (16k context, code-focused)

2. Basic Usage

# Ask a question (uses 'smart' profile by default)
cdc ask "explain asyncio in Python"

# Use a specific model profile
cdc ask -m fast "quick question"  # Uses Haiku (fast & cheap)
cdc ask -m powerful "complex task"  # Uses Opus 4 (most capable)

# With file context
cdc ask -f mycode.py "review this code"

# Pipe input
cat error.log | cdc ask "what's causing this error?"

# Interactive mode
cdc interactive

# Use specific API
cdc ask -a client "generate tests for this function"

3. Model Profile Management (v0.10.0+)

# List available model profiles
cdc model list
# ┌───────────┬──────────────────────────┬────────┬────────┐
# │ Profile   │ Model ID                 │ Input  │ Output │
# │ fast      │ claude-3-5-haiku-...     │ $0.80  │ $4.00  │
# │ smart     │ claude-sonnet-4-...      │ $3.00  │ $15.00 │ ← default
# │ powerful  │ claude-opus-4-...        │ $15.00 │ $75.00 │
# └───────────┴──────────────────────────┴────────┴────────┘

# Add a custom profile
cdc model add lightning claude-3-5-haiku-20241022 \
  --input-price 0.80 --output-price 4.00 \
  --description "Ultra-fast model for simple tasks"

# Set default model profile
cdc model set-default fast  # Use Haiku by default

# Set default for specific API config
cdc model set-default powerful --api enterprise  # Opus for enterprise API

# Show profile details
cdc model show smart

# Remove a profile
cdc model remove lightning

# Use profiles in any command
cdc ask -m fast "simple question"
cdc review -m powerful complex_file.py  # More thorough review
cdc generate tests -m smart mymodule.py  # Balanced approach

3. Code Generation Commands (NEW in v0.12.0, enhanced v0.13.0)

# Generate code from specification (single file)
cdc generate code --description "REST API client for weather data" -o client.py
cdc generate code --file spec.md -o implementation.go
cdc generate code --pdf requirements.pdf -o app.js
cdc generate code --url https://example.com/api-spec -o service.py

# Generate multi-file projects (NEW in v0.13.0)
cdc generate code --description "FastAPI REST API with auth" -o my-api/
cdc generate code --file spec.md -o project/ --dry-run  # Preview first
cdc generate code --file spec.md -o project/ --yes  # No confirmation
# Creates complete directory structure with multiple files

# Generate code with interactive refinement
cdc generate code --description "Database ORM" -o orm.py --interactive

# Generate code with project context
cdc generate code --file spec.md -o service.py --auto-context

# Add features to existing project (NEW: multi-file output in v0.13.0)
cdc generate feature --description "Add user authentication with JWT" src/
cdc generate feature --file feature-spec.md
cdc generate feature --pdf product-requirements.pdf --preview
cdc generate feature --url https://example.com/feature-spec

# Preview feature changes before applying
cdc generate feature --description "Add caching layer" src/ --preview
cdc generate feature --description "Add REST API" --dry-run  # Preview changes
cdc generate feature --file spec.md --yes  # Apply without confirmation

# Interactive feature implementation
cdc generate feature --description "Add logging" src/ --interactive

# Hunk-by-hunk approval (like git add -p) - NEW in v0.13.1
cdc generate feature -f spec.md
# At confirmation prompt, type 'patch' to review changes hunk-by-hunk
# Options: y (yes), n (no), s (skip file), q (quit), help

3.1 Interactive Diff Approval (v0.13.1+)

When applying file modifications, you can review and approve changes hunk-by-hunk, similar to git add -p:

# After generating feature/refactor changes:
cdc generate feature -f spec.md

# At the confirmation prompt:
Continue? (Y/n/preview/patch/help) patch

# For each file:
File: src/main.py
Modify file (3 hunk(s))

Hunk 1/3:
[Shows diff with syntax highlighting]
@@ -10,3 +10,5 @@
 def main():
-    print("old")
+    print("new")
+    logging.info("Started")

Apply this hunk? (y/n/s=skip file/q=quit/help) y  # Approve this hunk

Hunk 2/3:
[Shows next diff]
Apply this hunk? (y/n/s=skip file/q=quit/help) n  # Skip this hunk

Hunk 3/3:
[Shows final diff]
Apply this hunk? (y/n/s=skip file/q=quit/help) s  # Skip remaining in file

# File operations options:
Create this file? (y/n/s=skip/q=quit) y  # For new files
Delete this file? (y/n/s=skip/q=quit) n  # For file deletions

Options:

• y, yes - Apply this hunk/file
• n, no - Skip this hunk (keeps original)
• s, skip - Skip remaining hunks in current file
• q, quit - Stop reviewing and apply approved changes so far
• edit - Open files in $EDITOR before applying
• save - Save to custom location
• help - Show help message

Benefits:

• Fine-grained control over changes
• Keep original code for some hunks while applying others
• Syntax-highlighted diffs for easy review
• Edit files before applying for manual tweaks
• Save to custom location without -o flag
• Safe: only approved hunks are written

3.2 Edit and Save Options (v0.13.2+)

Before applying changes, you can edit files or save to custom locations:

Edit in $EDITOR

cdc generate feature -f spec.md

# At confirmation:
Continue? (Y/n/preview/patch/edit/save/help) edit

# Opens each file in your $EDITOR (vi, nano, code, etc.)
# Make manual adjustments, save and close
# Changes are applied after editing

Save to Custom Location

cdc generate code -d "REST API" -o /tmp/output

# At confirmation:
Continue? (Y/n/preview/patch/edit/save/help) save

# Single file:
Filename: my-custom-name.py  # Save to custom filename

# Multiple files:
Directory: /path/to/output/  # Save entire project elsewhere

Use Cases:

• Edit: Make manual tweaks before applying (fix formatting, adjust logic)
• Save: Try changes elsewhere before applying to project
• Edit + Preview: Review, edit, then apply with confidence
• Save for later: Generate code, save it, review offline, apply manually

Environment Variables:

• $EDITOR: Your preferred editor (e.g., export EDITOR=nano)
• Defaults to vi if $EDITOR not set

4. Developer Commands

# Generate tests (single file)
cdc generate tests mymodule.py -o tests/test_mymodule.py

# Generate tests for multiple files (NEW in v0.11.0)
cdc generate tests file1.py file2.py file3.py
cdc generate tests src/ --max-files 10

# Generate tests with interactive refinement
cdc generate tests mymodule.py --interactive

# Generate tests with context (includes dependencies, related files) - NEW in v0.8.1
cdc generate tests mymodule.py --auto-context

# Code review (single file)
cdc review mymodule.py

# Code review multiple files (NEW in v0.11.0)
cdc review file1.py file2.py file3.py
cdc review src/  # Review entire directory
cdc review  # Auto-detect git changes (staged → modified → current dir)

# Code review with auto-context (includes git, dependencies, tests)
cdc review mymodule.py --auto-context

# Code review with interactive follow-up questions
cdc review mymodule.py --interactive

# Review git changes (NEW in v0.11.0)
cdc git review --staged  # Review only staged changes
cdc git review --branch main..HEAD  # Review branch changes
cdc git review  # Review all modified files

# Debug errors with intelligent error parsing
python script.py 2>&1 | cdc debug --auto-context

# Generate documentation (single file)
cdc generate docs mymodule.py

# Generate docs for multiple files (NEW in v0.11.0)
cdc generate docs file1.py file2.py file3.py
cdc generate docs src/ --max-files 10

# Generate docs with interactive refinement
cdc generate docs mymodule.py --interactive

# Generate docs with context (includes dependencies) - NEW in v0.8.1  
cdc generate docs mymodule.py --auto-context

# Refactor (single file)
cdc refactor legacy_code.py

# Refactor multiple files (NEW in v0.11.0, enhanced v0.13.0)
cdc refactor file1.py file2.py file3.py
cdc refactor src/
cdc refactor  # Auto-detect git changes

# Multi-file refactoring with preview (NEW in v0.13.0)
cdc refactor src/ --dry-run  # Preview changes
cdc refactor src/ --yes  # Apply without confirmation
cdc refactor src/ --preview  # Review before applying

# Refactor with context (includes related files)
cdc refactor legacy_code.py --auto-context

# Refactor with interactive refinement
cdc refactor legacy_code.py --interactive

# Git commit message
git add .
cdc git commit

# Git commit message with context (includes history, branch) - NEW in v0.8.1
git add .
cdc git commit --auto-context

5. Context-Aware Operations (v0.8.0+)

# Auto-context includes: git info, dependencies, related files

# Review with full project context
cdc review mymodule.py --auto-context
# ✓ Context gathered (git, dependencies, tests)

# Debug with parsed error details (multi-language support)
python broken.py 2>&1 | cdc debug -f broken.py --auto-context
node app.js 2>&1 | cdc debug --auto-context  # JavaScript/TypeScript
go run main.go 2>&1 | cdc debug --auto-context  # Go
# ✓ Context gathered (error details, git context)
# Supports: Python, JavaScript, TypeScript, Go, Rust, Java

# Preview context before making API calls - NEW in v0.8.2
cdc context summary mymodule.py
# Shows: files, sizes, lines, estimated tokens, truncation warnings

# Ask questions with file context
cdc ask -f mycode.py --auto-context "how can I improve this?"
# ✓ Context gathered

# Refactor with related files
cdc refactor app.py --auto-context
# Automatically includes imported modules and dependencies

6. Custom Templates

# List all templates (built-in and user)
cdc template list

# Show template details
cdc template show code-review

# Add a custom template
cdc template add my-review -c "Review this code for {{focus}}: {{code}}" \
  -d "Custom review template" --category review

# Use a template (interactive variable input)
cdc template use debug-error

# Delete a user template
cdc template delete my-review

# Filter by category
cdc template list --category review

# Show only user templates
cdc template list --user

Built-in Templates

• code-review: Comprehensive code review with customizable focus
• code-review-security: Security-focused code review
• test-strategy: Generate testing strategy and test cases
• debug-error: Debug error with context
• optimize-performance: Performance optimization analysis
• refactor-clean: Clean code refactoring
• explain-code: Detailed code explanation
• api-design: API design assistance

6. Conversation History & Summarization (NEW in v0.8.3)

# List recent conversations
cdc history list

# Search conversations
cdc history list --search "python decorators"

# Export conversation
cdc history export 20240109_143022 -o chat.md

# Summarize to reduce token usage
cdc history summarize --latest
cdc history summarize 20240109_143022 --keep-recent 6

# Delete old conversations
cdc history delete 20240109_143022

Auto-Summarization in interactive mode:

• Automatically triggers when conversation exceeds 8,000 tokens
• Keeps recent messages (default: 4 pairs), summarizes older ones
• Reduces API costs by ~30-50% in long conversations
• Shows token savings after summarization

7. Workflows - AI-Powered Automation (v0.7.0+, Enhanced v0.16.1)

Chain multiple AI operations into automated sequences with YAML workflows.

Basic Workflow Usage

# Run a workflow
cdc workflow run my-workflow.yaml

# Pass variables
cdc workflow run workflow.yaml --var file=main.py --var api=local

# List available workflows
cdc workflow list

# Show workflow details
cdc workflow show workflow.yaml

Example: Simple Review & Test Workflow

name: "Review and Test"
description: "Review code and generate tests"

steps:
  - name: code-review
    command: review
    args:
      file: "{{target_file}}"
      api: "local"  # Use FREE local AI
      model: "smart-local"
    output_var: review_result
  
  - name: generate-tests
    command: generate tests
    args:
      file: "{{target_file}}"
      api: "local"
      model: "code-local"
    output_var: tests
  
  - name: run-tests
    shell: "pytest tests/"
    continue_on_error: true

Run it:

cdc workflow run review-test.yaml --var target_file=app.py

Example: AI Decision Making (NEW in v0.16.1)

Use the ask command to make intelligent workflow decisions:

name: "Adaptive Code Analysis"
description: "AI analyzes code complexity and adapts workflow"

steps:
  # AI rates complexity
  - name: check-complexity
    command: ask
    args:
      prompt: "Rate this code complexity 1-10, respond with number only: {{code}}"
      api: "local"  # FREE
      model: "fast-local"
    output_var: complexity
  
  # Conditional deep review for complex code
  - name: deep-review
    command: review
    args:
      file: "{{file}}"
      api: "personal"  # Use paid API for complex code
      model: "powerful"
    if: "{{complexity}} > 7"
    approval_required: true  # Ask before using paid API
  
  # Simple refactor for medium complexity
  - name: simple-refactor
    command: refactor
    args:
      file: "{{file}}"
      api: "local"  # FREE
    if: "{{complexity}} >= 5 and {{complexity}} <= 7"

Key Features:

• ✨ ask command for AI queries
• 💰 Mix free local and paid cloud AI
• 🎯 Conditional execution based on AI responses
• ✋ Approval gates before expensive operations

Example: Cost-Optimized Workflow

name: "Smart Cost Optimization"
description: "Use local AI first, cloud only when critical"

steps:
  # Quick scan with FREE local AI
  - name: quick-scan
    command: ask
    args:
      prompt: "Quick scan for obvious issues: {{code}}"
      api: "local"  # FREE
      model: "fast-local"
    output_var: issues
  
  # Only use expensive cloud AI if critical issues found
  - name: deep-security-analysis
    command: ask
    args:
      prompt: "Deep security analysis: {{code}}"
      api: "personal"  # PAID
      model: "powerful"  # Claude Opus - expensive
      system: "You are a security expert"
    if: "{{issues}} contains 'critical' or {{issues}} contains 'security'"
    approval_required: true
  
  # Tests with local code-specialized model (FREE)
  - name: generate-tests
    command: generate tests
    args:
      file: "{{file}}"
      api: "local"  # FREE
      model: "code-local"

Cost Savings:

• Local AI for initial scans: $0
• Cloud AI only when needed: saves 70-90%
• Approval gates prevent accidental costs

Workflow Step Types

1. Command Steps (cdc commands)

- name: review-code
  command: review  # or: generate tests, debug, refactor, etc.
  args:
    file: "{{file}}"
    api: "local"
    model: "smart-local"
  output_var: result

Supported commands: review, generate tests, generate docs, refactor, debug, git commit, ask (v0.16.1)

2. Shell Steps (any command)

- name: run-tests
  shell: "pytest tests/ -v"
  output_var: test_output
  continue_on_error: true

3. Set Steps (variables)

- name: set-config
  set: api_endpoint
  value: "https://api.example.com"

Advanced Features

Conditional Execution:

if: "{{score}} > 7"  # Simple comparison
if: "{{result}} contains 'error'"  # String matching
if: "{{tests.success}}"  # Access step results

Approval Gates:

approval_required: true  # Pause for user confirmation

Error Handling:

continue_on_error: true  # Don't stop workflow on failure

Variable Interpolation:

{{variable}}  # Simple variable
{{step_name.output}}  # Step output
{{step_name.success}}  # Step success status

Example Workflows Included

Location: examples/workflows/

1. review-and-refactor.yaml - Basic review workflow
2. ai-decision-workflow.yaml - AI-powered decisions (v0.16.1)
3. multi-model-workflow.yaml - Cost optimization (v0.16.1)

8. Usage Tracking

# View all usage
cdc usage

# Last 7 days
cdc usage --days 7

# Filter by API
cdc usage --api client

9. TOON Format (Optional - Reduces Tokens by 30-60%)

# Check if TOON is installed
cdc toon info

# Convert JSON to TOON
echo '{"users": [{"id": 1, "name": "Alice"}]}' | cdc toon encode
# Output:
# users[1]{id,name}:
# 1,Alice

# Convert file
cdc toon encode data.json -o data.toon

# Convert TOON back to JSON
cdc toon decode data.toon -o data.json

# Use in workflows
cat large_data.json | cdc toon encode | cdc ask "analyze this data"

Configuration

Secure API Key Storage

🔐 Your API keys are stored securely and never in plain text.

• macOS: Keychain
• Linux: Secret Service API (GNOME Keyring, KWallet)
• Windows: Windows Credential Locker
• Fallback: Encrypted file (if keyring unavailable)

Keys are automatically migrated from plaintext on first run. You can also manually migrate:

cdc config migrate-keys

Global Configuration

Configuration metadata is stored in ~/.claude-dev-cli/config.json (API keys are NOT in this file):

{
  "api_configs": [
    {
      "name": "personal",
      "api_key": "",  // Empty - actual key in secure storage
      "description": "My personal API key",
      "default": true
    },
    {
      "name": "client",
      "api_key": "",  // Empty - actual key in secure storage
      "description": "Client's Enterprise API",
      "default": false
    }
  ],
  "default_model": "claude-3-5-sonnet-20241022",
  "max_tokens": 4096
}

Project-Specific Configuration

Create .claude-dev-cli in your project root:

{
  "name": "My Project",
  "api_config": "client",
  "system_prompt": "You are a Python expert focused on data science.",
  "allowed_commands": ["all"]
}

The tool will automatically use the client's API when you run commands in that project!

API Routing Strategy

Scenario: Client Work with Enterprise API

# 1. Set up client's API
export CLIENT_ANTHROPIC_API_KEY="sk-ant-..."
cdc config add client --description "Client Enterprise API"

# 2. Create project config
cd /path/to/client/project
cat > .claude-dev-cli << 'EOF'
{
  "name": "Client Project",
  "api_config": "client",
  "system_prompt": "You are a senior Python developer."
}
EOF

# 3. All commands in this directory now use client's API
cdc generate tests app.py  # Uses client's API
cdc review code.py         # Uses client's API

# 4. Your personal projects use your API
cd ~/my-personal-project
cdc generate tests app.py  # Uses your personal API

What the Client Can See

When using a client's Enterprise API:

✅ They CAN see:

• Total API calls and token usage
• Costs associated with the API key
• Potentially conversation history (if enabled by admin)

❌ They CANNOT see:

• Which machine you used
• Your other API keys or projects
• Local files unless you sent the content

Command Reference

Core Commands

Command	Description
cdc ask <prompt>	Ask Claude a question
cdc interactive	Start interactive chat
cdc usage	Show usage statistics

Configuration

Command	Description
cdc config add <name>	Add API configuration
cdc config list	List all API configs

Code Generation

Command	Description
cdc generate tests <file>	Generate pytest tests
cdc generate docs <file>	Generate documentation

Code Analysis

Command	Description
cdc review <file>	Review code
cdc debug -f <file> -e <error>	Debug code
cdc refactor <file>	Refactoring suggestions

Templates

Command	Description
cdc template list	List all templates
cdc template show <name>	Show template details
cdc template add <name>	Add new template
cdc template delete <name>	Delete user template
cdc template use <name>	Use template interactively

Git Helpers

Command	Description
cdc git commit	Generate commit message

Workflows (v0.7.0+)

Command	Description
cdc workflow run <file>	Execute workflow from YAML
cdc workflow list	List available workflows
cdc workflow show <file>	Show workflow details

Ollama (Local AI) (v0.16.0+)

Command	Description
cdc ollama list	List local models
cdc ollama pull <model>	Download model
cdc ollama show <model>	Show model details

Model Profiles (v0.10.0+)

Command	Description
cdc model list	List all model profiles
cdc model show <name>	Show profile details
cdc model add <name> <model_id>	Add custom profile
cdc model set-default <name>	Set default profile

Options

Common Options

• -a, --api <name>: Use specific API config
• -m, --model <name>: Use specific Claude model
• -s, --system <prompt>: Set system prompt
• -f, --file <path>: Include file in prompt
• -o, --output <path>: Save output to file

Model Profiles

Anthropic (Claude):

• fast / claude-3-5-haiku-20241022 - Fast & economical ($0.80/$4 per Mtok)
• smart / claude-sonnet-4-5-20250929 - Balanced (default) ($3/$15 per Mtok)
• powerful / claude-opus-4-20250514 - Most capable ($15/$75 per Mtok)

OpenAI (GPT): (requires pip install 'claude-dev-cli[openai]')

• fast-openai / gpt-3.5-turbo - Fast & cheap ($0.50/$1.50 per Mtok)
• smart-openai / gpt-4-turbo - Balanced ($10/$30 per Mtok)
• powerful-openai / gpt-4 - High capability ($30/$60 per Mtok)

Ollama (Local - FREE): (requires pip install 'claude-dev-cli[ollama]')

• fast-local / mistral - Fast inference (8k context, FREE)
• smart-local / mixtral - Powerful reasoning (32k context, FREE)
• code-local / codellama - Code specialist (16k context, FREE)

Usage:

# Use model profile
cdc ask -m fast "quick question"  # Uses Haiku
cdc ask -m fast-local "question"  # Uses Mistral (FREE)

# Use specific model ID
cdc ask -m claude-opus-4-20250514 "complex task"
cdc ask -m gpt-4-turbo "openai task"

Examples

Test Generation

# Generate tests for a module
cdc generate tests src/calculator.py -o tests/test_calculator.py

# Review the tests
cdc review tests/test_calculator.py

Code Review Workflow

# Review before committing
cdc review mycode.py

# Fix issues, then generate commit message
git add mycode.py
cdc git commit

Debugging

# Run script and analyze errors
python buggy_script.py 2>&1 | cdc debug

# Or debug a specific file
cdc debug -f buggy_script.py -e "NameError: name 'x' is not defined"

Using Multiple APIs

# Personal work
cdc ask "explain decorators" -a personal

# Client work
cdc generate tests client_code.py -a client

# Check usage per API
cdc usage --api personal
cdc usage --api client

Project Management & Automation (v0.17.0+)

Automated Ticket Execution

# Create a ticket
cdc tickets create "Implement user authentication API" \
  --description "JWT-based auth with email/password" \
  --priority high \
  --type feature

# Output: Created ticket: TASK-A1B2C3D4

# Execute the ticket with full AI automation
cdc ticket execute TASK-A1B2C3D4 --commit --notify

# What happens:
# 1. 📥 Fetches ticket and analyzes requirements
# 2. 🧠 AI generates implementation plan
# 3. 💻 Generates api/auth.py with login endpoint
# 4. ✅ Generates tests/test_auth.py with all acceptance criteria
# 5. 📝 Commits: "feat(TASK-A1B2C3D4): Implement user authentication API"
# 6. 🎫 Updates ticket status to "completed"
# 7. 📊 Logs to .cdc-logs/progress.md
# 8. 🔔 Sends push notification

# Review progress
cdc log report

Bug Reporting with AI Triage

# Report a bug - AI automatically classifies it
cdc bug report \
  --title "App crashes when email contains + character" \
  --description "Users with + in email address cannot login" \
  --expected "All valid email addresses should work" \
  --actual "Server returns 500 error for emails with +" \
  --steps "Enter email [e-mail]" \
  --steps "Click login" \
  --steps "App crashes" \
  --environment production

# AI automatically:
# ✅ Severity: HIGH (functionality broken)
# ✅ Category: FUNCTIONALITY
# ✅ Priority: high
# ✅ Creates ticket: BUG-XYZ123
# ✅ Checks for duplicates
# ✅ Sends notification if critical

# List all bugs
cdc tickets list --status open

# Execute bug fix
cdc ticket execute BUG-XYZ123 --commit

Progress Tracking

# Initialize logging for a project
cdc log init "E-Commerce Platform"

# Add manual log entries
cdc log entry "Started authentication module" --level info
cdc log entry "Completed user registration" --ticket TASK-123 --level success
cdc log entry "Found issue with password validation" --level warning

# Generate full progress report
cdc log report

# Output:
# # Progress Summary
# 
# **Total Entries:** 15
# **Successes:** ✅ 8
# **Errors:** ❌ 1
# 
# ## ✅ 2026-02-01 02:30:15
# **Ticket:** TASK-123
# Completed user registration
# ...

Using with repo-tickets Backend

# Install repo-tickets integration
pip install 'claude-dev-cli[all-pm]'

# Navigate to your project with repo-tickets
cd my-project
tickets init  # Initialize repo-tickets if needed

# Create a ticket in repo-tickets
tickets create "Add payment gateway integration"
# Output: Created TICKET-456

# Execute it with AI
cdc ticket execute TICKET-456 --backend repo-tickets --commit --notify

# AI will:
# - Fetch ticket from repo-tickets
# - Generate payment gateway code
# - Generate tests
# - Update repo-tickets status
# - Commit to your repo

Cost Optimization with Local AI

# Use FREE local AI (Ollama) for code generation
# Install Ollama first: https://ollama.ai

# Pull a code model
cdc ollama pull codellama

# Execute ticket with local AI (zero cost!)
cdc ticket execute TASK-789 --api local --commit

# Result: Code generated with $0.00 cost!

Notifications Setup

# Test notifications (uses ntfy.sh - FREE, no signup!)
cdc notify test --topic my-project

# Check your phone or browser at: https://ntfy.sh/my-project
# You'll see: "✅ claude-dev-cli notifications are working!"

# All ticket executions can send notifications:
cdc ticket execute TASK-999 --commit --notify

# Critical bugs auto-notify:
cdc bug report --title "Database connection fails" \
  --description "Cannot connect to production DB" \
  --expected "Should connect" \
  --actual "Connection timeout" \
  --environment production
# 🚨 Sends URGENT notification automatically

Complete Workflow Example

# 1. Setup
pip install 'claude-dev-cli[all-pm]'
cdc log init "My Awesome Project"

# 2. Create tickets for your sprint
cdc tickets create "User registration" --priority high
cdc tickets create "User login" --priority high
cdc tickets create "Password reset" --priority medium
cdc tickets create "Email verification" --priority low

# 3. Execute tickets one by one
cdc ticket execute TASK-001 --commit --notify
cdc ticket execute TASK-002 --commit --notify

# 4. Bug is reported
cdc bug report \
  --title "Registration fails for Gmail users" \
  --description "Gmail addresses rejected" \
  --expected "Should accept all emails" \
  --actual "Validation error"

# 5. Fix the bug
cdc ticket execute BUG-001 --commit --notify

# 6. Review progress
cdc log report
cdc tickets list

# 7. All done! 🎉
# - Code generated
# - Tests written
# - Tickets updated
# - Commits made
# - Team notified
# - Progress logged

Environment Variables

Variable	Description
{NAME}_ANTHROPIC_API_KEY	API key for config named "name"
ANTHROPIC_API_KEY	Default API key

Usage Log Format

Usage logs are stored in ~/.claude-dev-cli/usage.jsonl:

{
  "timestamp": "2024-12-28T03:00:00.000000",
  "api_config": "client",
  "model": "claude-3-5-sonnet-20241022",
  "prompt_preview": "Generate tests for...",
  "input_tokens": 1523,
  "output_tokens": 2847,
  "duration_ms": 3421
}

Development

# Clone the repository
git clone https://github.com/thinmanj/claude-dev-cli
cd claude-dev-cli

# Install in development mode
pip install -e ".[dev]"

# Run tests
pytest

# Format code
black src/
ruff check src/

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT License - see LICENSE for details.

Author

Julio (@thinmanj)

Acknowledgments

• Built with Anthropic's Claude API
• CLI powered by Click
• Beautiful terminal output with Rich

Support

• 🐛 Report bugs
• 💡 Request features
• 📖 Documentation