Major documentation update focused on technical accuracy and developer clarity: Documentation Changes: - Rewrote README.md with focus on hooks system architecture - Updated all core docs (Overview, Integration, Performance) to match implementation - Created 6 missing configuration docs for undocumented YAML files - Updated all 7 hook docs to reflect actual Python implementations - Created docs for 2 missing shared modules (intelligence_engine, validate_system) - Updated all 5 pattern docs with real YAML examples - Added 4 essential operational docs (INSTALLATION, TROUBLESHOOTING, CONFIGURATION, QUICK_REFERENCE) Key Improvements: - Removed all marketing language in favor of humble technical documentation - Fixed critical configuration discrepancies (logging defaults, performance targets) - Used actual code examples and configuration from implementation - Complete coverage: 15 configs, 10 modules, 7 hooks, 3 pattern tiers - Based all documentation on actual file review and code analysis Technical Accuracy: - Corrected performance targets to match performance.yaml - Fixed timeout values from settings.json (10-15 seconds) - Updated module count and descriptions to match actual shared/ directory - Aligned all examples with actual YAML and Python implementations The documentation now provides accurate, practical information for developers working with the Framework-Hooks system, focusing on what it actually does rather than aspirational features. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
5.5 KiB
Installation Guide
Framework-Hooks provides intelligent session management for Claude Code through Python hooks that run at specific lifecycle events.
Prerequisites
- Python 3.8+ (Python 3.x required by hook scripts)
- Claude Code application
- Write access to your system's hook installation directory
Installation Steps
1. Verify Python Installation
python3 --version
# Should show Python 3.8 or higher
2. Clone or Extract Framework-Hooks
Place the Framework-Hooks directory in your SuperClaude installation:
YourProject/
├── SuperClaude/
│ └── Framework-Hooks/ # This repository
└── other-files...
3. Install Hook Scripts
Framework-Hooks includes pre-configured hook registration files:
settings.json- Claude Code hook configurationsuperclaude-config.json- SuperClaude framework settings
These files configure 7 hooks to run at specific lifecycle events:
session_start.py- Session initialization (<50ms target)pre_tool_use.py- Tool preparation (<200ms target)post_tool_use.py- Tool usage recording (<100ms target)pre_compact.py- Context compression (<150ms target)notification.py- Notification handling (<50ms target)stop.py- Session cleanup (<100ms target)subagent_stop.py- Subagent coordination (<100ms target)
4. Directory Structure Verification
After installation, verify this structure exists:
Framework-Hooks/
├── hooks/
│ ├── session_start.py
│ ├── pre_tool_use.py
│ ├── post_tool_use.py
│ ├── pre_compact.py
│ ├── notification.py
│ ├── stop.py
│ ├── subagent_stop.py
│ └── shared/ # 9 shared modules
│ ├── framework_logic.py
│ ├── compression_engine.py
│ ├── learning_engine.py
│ ├── mcp_intelligence.py
│ ├── pattern_detection.py
│ ├── intelligence_engine.py
│ ├── logger.py
│ ├── yaml_loader.py
│ └── validate_system.py
├── config/ # 12+ YAML configuration files
│ ├── session.yaml
│ ├── performance.yaml
│ ├── compression.yaml
│ ├── modes.yaml
│ ├── mcp_orchestration.yaml
│ ├── orchestrator.yaml
│ ├── logging.yaml
│ └── validation.yaml
├── patterns/ # 3-tier pattern system
│ ├── minimal/ # Basic patterns (3-5KB each)
│ ├── dynamic/ # Feature-specific (8-12KB each)
│ └── learned/ # User adaptations (10-20KB each)
├── cache/ # Runtime cache directory
└── docs/ # Documentation
5. Configuration Check
The system ships with conservative defaults:
- Logging: Disabled by default (
logging.yamlhasenabled: false) - Performance targets: session_start <50ms, pre_tool_use <200ms
- Timeouts: 10-15 seconds per hook execution
- All hooks enabled: via settings.json configuration
6. Test Installation
Run the validation system to verify installation:
cd Framework-Hooks/hooks/shared
python3 validate_system.py --check-installation
This will verify:
- Python dependencies available
- All hook files are executable
- Configuration files are valid YAML
- Required directories exist
- Shared modules can be imported
Verification
Quick Test
Start a Claude Code session. You should see hook execution if logging is enabled (edit config/logging.yaml to enable logging).
Check Hook Registration
The hooks register automatically through:
settings.json- Defines 7 hooks with 10-15 second timeouts- Commands like
python3 ~/.claude/hooks/session_start.py - Universal matcher
"*"applies to all sessions
Performance Verification
Hooks should execute within performance targets:
- Session start: <50ms
- Pre tool use: <200ms
- Post tool use: <100ms
- Other hooks: <100-150ms each
Configuration
Default Settings
All hooks start enabled with conservative defaults:
- Logging disabled (
config/logging.yaml) - Error-only log level
- 30-day log retention
- Privacy-safe logging (sanitizes user content)
Enable Logging (Optional)
To see hook activity, edit config/logging.yaml:
logging:
enabled: true
level: "INFO" # or DEBUG for verbose output
Pattern System
The 3-tier pattern system loads automatically:
- minimal/: Basic project detection (always loaded)
- dynamic/: Feature-specific patterns (loaded on demand)
- learned/: User-specific adaptations (evolve with usage)
Next Steps
After installation:
- Normal Usage: Start Claude Code sessions normally - hooks run automatically
- Monitor Performance: Check that hooks execute within target times
- Review Logs: Enable logging to see hook decisions and learning
- Customize Patterns: Add project-specific patterns to
patterns/directories
What Framework-Hooks Does
Once installed, the system automatically:
- Detects project type and loads appropriate patterns
- Activates relevant modes and MCP servers based on context
- Applies learned preferences from previous sessions
- Optimizes performance based on resource constraints
- Learns from usage patterns to improve future sessions
The system operates transparently - no manual invocation required.
Troubleshooting
If you encounter issues, see TROUBLESHOOTING.md for common problems and solutions.