Claude-Code-Workflow/.claude/workflows/python-tools-strategy.md

name, description, type

name	description	type
python-tools-strategy	Command reference for Python-based tool invocation	command-reference

Python Tools Command Reference

⚡ Quick Commands

Smart Analysis: pycli --analyze --query "search term" --tool [gemini/codex] Direct Tool Invocation: pycli --analyze --tool [gemini/codex] -p "prompt" Vector Database Setup: pycli --init Vector Database Update: pycli --update-embeddings

⏰ When to Use What

🔄 Vector Database Timing

# FIRST TIME (run once per project)
pycli --init

# DAILY (when files change)
pycli --update-embeddings

# BEFORE ANALYSIS (check status)
pycli --status

🎯 Tool Selection Timing

• Code Discovery → Use pycli --analyze --query to find relevant files
• Direct Analysis → Use pycli --analyze -p when you know what to analyze
• Development → Use --tool codex for implementation tasks
• Understanding → Use --tool gemini for analysis and exploration

🎯 Core Commands

Smart Analysis (Recommended)

# Find similar code patterns and analyze
pycli --analyze --query "authentication patterns" --tool gemini

# Search with development context
pycli --analyze --query "error handling" --tool codex

# Both discovery and analysis
pycli --analyze --query "database connections" --tool both

Direct Tool Invocation

# Direct analysis with known context
pycli --analyze --tool gemini -p "analyze authentication patterns"

# Direct development task
pycli --analyze --tool codex -p "implement user login"

# Status and testing
pycli --status
pycli --test-search

Vector Database Operations

# Initial setup (run once per project)
pycli --init

# Daily updates (run when files change)
pycli --update-embeddings

# Status check
pycli --status

📊 Command Matrix

What You Want	Command	Use Case
Smart analysis	pycli --analyze --query "pattern" --tool gemini	Code discovery & analysis
Direct analysis	pycli --analyze --tool gemini -p "prompt"	Known target analysis
Generate code	pycli --analyze --tool codex -p "task"	Development
Setup project	pycli --init	First time setup
Update search index	pycli --update-embeddings	Maintenance
Check status	pycli --status	System health

🚀 Usage Examples

Replace Gemini Wrapper

# OLD: ~/.claude/scripts/gemini-wrapper -p "analyze auth patterns"
# NEW: pycli --analyze --tool gemini -p "analyze auth patterns"

Replace Codex Commands

# OLD: codex --full-auto exec "implement login"
# NEW: pycli --analyze --tool codex -p "implement login"

Smart Context Discovery

# Find relevant files first, then analyze
pycli --analyze --query "user authentication" --tool gemini

# Results include:
# - Hierarchical vector database search
# - Semantically similar files from project and parent directories
# - Generated tool command with intelligent context
# - Executed analysis with smart file selection

🔧 Command Options

pycli (Unified Interface)

pycli [command] [options]

Commands:
  --init                Initialize vector database for current project
  --analyze             Run analysis with AI tools
  --status              Show system status and health
  --test-search         Test vector search functionality
  --update-embeddings   Update vector embeddings for changed files

Analysis Options:
  --tool [gemini|codex|both]   Which AI tool to use (default: gemini)
  -p, --prompt TEXT            Direct prompt for analysis
  --query TEXT                 Semantic search query for context discovery
  --top-k INTEGER              Number of similar files to find (default: 10)
  --similarity-threshold FLOAT Minimum similarity score (0.0-1.0)

Output Options:
  --quiet                      Suppress progress output
  --verbose                    Show detailed analysis information
  --output [patterns|json]     Output format (default: patterns)

Installation & Setup

# Install pycli system
bash D:/Claude_dms3/.claude/scripts/install_pycli.sh

# The script will automatically add ~/.claude/scripts/ to your PATH
# Reload your shell configuration
source ~/.bashrc  # or ~/.zshrc

# Verify installation - now you can use pycli directly
pycli --help

📋 Common Workflows

🚀 First-Time Setup (Vector Database)

# 1. Install pycli system
bash D:/Claude_dms3/.claude/scripts/install_pycli.sh

# 2. Initialize vector database for project
cd /path/to/your/project
pycli --init

# 3. Verify setup works
pycli --status

# 4. Test search functionality
pycli --test-search

🎯 Analysis Workflow (Recommended)

# 1. Update vectors (if files changed)
pycli --update-embeddings

# 2. Smart analysis with context discovery
pycli --analyze --query "what you're looking for" --tool gemini

# 3. Development with context
pycli --analyze --query "related patterns" --tool codex

⏰ When to Run Commands

🔄 Vector Database Maintenance

# WHEN: First time using system
pycli --init

# WHEN: Files have been added/modified (daily/after coding)
pycli --update-embeddings

# WHEN: Before starting analysis (check if system ready)
pycli --status

🎯 Analysis Timing

# WHEN: You need to find relevant code patterns
pycli --analyze --query "search term" --tool gemini

# WHEN: You have specific prompt and know context
pycli --analyze --tool gemini -p "specific prompt"

# WHEN: You want to develop/implement something
pycli --analyze --query "similar implementations" --tool codex

Integration with Existing Tools

# In place of gemini-wrapper
pycli --analyze --tool gemini -p "$YOUR_PROMPT"

# In place of codex commands
pycli --analyze --tool codex -p "$YOUR_TASK"

# Enhanced with hierarchical context discovery
pycli --analyze --query "relevant context" --tool both

🎯 Quick Reference

🚀 Most Common Commands

# 1. Smart analysis (recommended first choice)
pycli --analyze --query "what you're looking for" --tool gemini

# 2. Direct tool call (when you know exactly what to analyze)
pycli --analyze --tool codex -p "what you want to do"

# 3. Keep embeddings updated (run after file changes)
pycli --update-embeddings

⚙️ Configuration (config.yaml)

# Essential settings only
embeddings:
  enabled: true
  similarity_threshold: 0.3

tools:
  default_tool: "gemini"
  timeout: 300

🐛 Troubleshooting

# Check if everything works
pycli --status

# Rebuild if issues
pycli --init

# Test search functionality
pycli --test-search

🎪 Integration Decision Tree

Need to analyze code?
├─ Do you know specific files to analyze?
│  ├─ YES → Use: pycli --analyze --tool [gemini/codex] -p "prompt"
│  └─ NO → Use: pycli --analyze --query "search term" --tool [gemini/codex]
└─ Is vector database updated?
   ├─ UNSURE → Run: pycli --status
   ├─ NO → Run: pycli --update-embeddings
   └─ YES → Proceed with analysis

🏗️ Hierarchical Vector Database

Key Features

• Automatic Parent Discovery: Subdirectories automatically use parent's vector database
• No Redundant Vectorization: Avoids duplicate processing in project subdirectories
• Central Storage: All vector databases stored in ~/.claude/vector_db/
• Path-based Organization: Vector DBs organized by project directory structure

How It Works

# Project structure
/home/user/myproject/
├── src/
│   └── auth/          # Uses parent's vector DB
└── tests/             # Uses parent's vector DB

# Vector database structure
~/.claude/vector_db/
└── home_user_myproject/    # Single DB for entire project
    ├── embeddings.pkl
    └── index.json

Usage Examples

# Initialize at project root
cd /home/user/myproject
pycli --init

# Work in subdirectory (automatically finds parent DB)
cd src/auth
pycli --analyze --query "authentication patterns"  # Uses parent's DB

# Work in another subdirectory
cd ../../tests
pycli --analyze --query "test patterns"  # Uses same parent DB

🔧 Vector Database Setup & Maintenance

⚡ One-Time System Setup

# 1. Install dependencies (first time only)
cd .claude/python_script && pip install -r requirements.txt

# 2. Initialize vector database (creates embeddings)
python indexer.py --rebuild-index --update-embeddings

# 3. Verify setup works
python cli.py --status

# 4. Test search functionality
python cli.py --test-search

📋 What Happens During Setup

1. File Indexing: Scans project files and creates index
2. Model Download: Downloads AI model (first time only, ~500MB)
3. Embedding Generation: Creates vector representations of code
4. Cache Creation: Saves embeddings to .claude/cache/embeddings/

🎯 Verification Checklist

After setup, verify these work:

• python cli.py --status shows "System ready"
• python cli.py --test-search returns results
• Files exist: .claude/cache/embeddings/embeddings.pkl
• Search works: python analyzer.py --query "test"

🐛 Common Issues & Fixes

Nothing works / Setup failed

# Nuclear option - reset everything
rm -rf .claude/cache/embeddings/*
python indexer.py --rebuild-index --update-embeddings

Slow performance

# In config.yaml - reduce batch size
embeddings:
  batch_size: 16

No search results found

# In config.yaml - lower similarity threshold
embeddings:
  similarity_threshold: 0.1

Memory errors during setup

# In config.yaml - use smaller batches
embeddings:
  batch_size: 8

Model download fails

# Manual model download
python -c "from sentence_transformers import SentenceTransformer; SentenceTransformer('all-MiniLM-L6-v2')"

📋 Usage Rules & Best Practices

🎯 Core Rules

1. Always check status first - Run python cli.py --status before analysis
2. Update after file changes - Run indexer.py --update-embeddings when files modified
3. Use vector search for discovery - Use analyzer.py --query when exploring code
4. Use direct tools for known targets - Use cli.py --analyze for specific analysis
5. Prefer context-aware tools - Enhanced Python tools over legacy shell scripts

⏰ Maintenance Schedule

# DAILY (or after coding sessions)
python .claude/python_script/indexer.py --update-embeddings

# WEEKLY (or when config changes)
python .claude/python_script/cli.py --status  # Check system health

# MONTHLY (or after major project changes)
python .claude/python_script/indexer.py --rebuild-index --update-embeddings

🎯 Tool Selection Rules

Use cli.py --analyze --query when:

• ✅ Exploring unfamiliar codebase
• ✅ Looking for similar code patterns
• ✅ Need context discovery for complex tasks
• ✅ Want smart file selection for tool execution

Use cli.py --analyze -p when:

• ✅ You know exactly what files to analyze
• ✅ Direct prompt execution without context search
• ✅ Quick tool invocation with known targets

Use indexer.py when:

• ✅ First time setup
• ✅ Files have been added/modified
• ✅ System performance degraded
• ✅ Configuration changed

🚀 Migration from Legacy Tools

Replace gemini-wrapper

# OLD (shell-based)
~/.claude/scripts/gemini-wrapper -p "analyze authentication"

# NEW (Python-based with hierarchical vector context)
pycli --analyze --query "authentication" --tool gemini

Replace codex commands

# OLD (direct execution)
codex --full-auto exec "implement user login"

# NEW (context-aware development with hierarchical DB)
pycli --analyze --query "login implementation patterns" --tool codex

Integration workflow

1. Install pycli - Run installation script once
2. Initialize projects - Run pycli --init in each project root
3. Replace commands - Update scripts to use pycli instead of direct Python calls
4. Enjoy hierarchical benefits - Automatic parent DB discovery in subdirectories

🎉 Advanced Features

Bash Wrapper Benefits

• Unified Interface: Single pycli command for all operations
• Smart Path Detection: Automatically finds project roots and vector databases
• Environment Management: Configurable Python interpreter path
• Hierarchical Support: Intelligent parent directory discovery