Claude-Sync learns from your command patterns to provide intelligent optimization suggestions, safety warnings, and contextual assistance across distributed computing environments. The system learns successful patterns across hosts while maintaining military-grade privacy through encrypted, abstracted learning data.

# Install and activate
curl -sL https://raw.githubusercontent.com/drejom/claude-sync/main/bootstrap.sh | bash -s install
~/.claude/claude-sync/bootstrap.sh activate --global

# Verify installation
~/.claude/claude-sync/bootstrap.sh status

Intelligent Command Optimization:

• grep → rg suggestions with confidence scoring
• SLURM resource optimization based on historical success
• Container workflow recommendations
• Safety warnings for dangerous operations

Learning Intelligence:

• Cross-host pattern sharing through encrypted P2P mesh
• Workflow sequence recognition and optimization
• Adaptive schema evolution based on your usage patterns
• Context-aware assistance for bioinformatics, HPC, and R workflows

Zero Configuration Security:

• Hardware-based host identity (survives OS reinstalls)
• Daily automatic key rotation with cleanup
• Military-grade encryption (Fernet + PBKDF2)
• Zero repository contamination guarantee

• Hook execution: <10ms (real-time suggestions)
• Learning operations: <1ms overhead
• Memory usage: <50MB total
• Cross-platform: macOS, Linux, WSL

• Automatic command upgrades: grep → rg, find → fd
• AI learns your successful command patterns and suggests improvements
• HPC-specific optimizations for SLURM, Singularity, genomics tools
• Safety warnings for destructive operations

• Learns filesystem topology across your hosts securely
• Routes commands to optimal hosts based on data location and capabilities
• Bioinformatics workflow awareness (FASTQ, BAM, VCF file handling)
• Supports mixed Tailscale + SSH environments

• Monitors resource usage patterns (memory, CPU, execution time)
• Suggests optimal HPC resources based on workload history
• Cross-host performance data sharing with encrypted sync

• Comprehensive documentation review system with specialized agents
• Ruthless editor that detects AI/inexperienced author patterns
• Context-aware analysis for R packages, READMEs, vignettes
• Example validation with syntax and runtime testing
• Audience optimization for user/developer/scientist content

• Auto-updates hooks when improvements are available
• Documentation review before git pushes
• Secure cross-host learning data synchronization

~/.claude/claude-sync/bootstrap.sh status         # Show activation status
~/.claude/claude-sync/bootstrap.sh test          # Run system tests
~/.claude/claude-sync/bootstrap.sh deactivate    # Clean removal
~/.claude/claude-sync/bootstrap.sh rollback      # Emergency restore

Installation issues:

~/.claude/claude-sync/bootstrap.sh diagnostics --health-score

Performance problems:

~/.claude/claude-sync/bootstrap.sh test --performance

Complete reset:

~/.claude/claude-sync/bootstrap.sh deactivate --purge

Claude-Sync uses Claude Code hooks for real-time optimization and specialized agents for deep analysis:

• PreToolUse: Command optimization and safety warnings
• PostToolUse: Silent learning data collection
• UserPromptSubmit: Context injection from learned patterns

Learning data stays encrypted locally (~/.claude/learning/*.enc) and never touches the repository.

# You type:
grep "function" *.R

# Hook suggests:
🚀 Optimized command: rg "function" *.R

# You type:
ls /data/projects/

# Hook suggests (after learning):
🧠 Learned pattern detected
Similar commands were previously run on:
  • ssh gemini-data1 'ls /data/projects/' (85% match)

Each project can customize which hooks to enable in .claude/settings.local.json:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          {
            "type": "command",
            "command": "$HOME/.claude/claude-sync/hooks/bash-optimizer-enhanced.py"
          },
          {
            "type": "command",
            "command": "$HOME/.claude/claude-sync/hooks/ssh-router-enhanced.py"
          }
        ]
      }
    ]
  }
}

Hooks are installed globally but can be disabled per-project by not including them in the project's settings.

The SSH router learns your infrastructure:

• Filesystem topology: Which paths exist on which hosts
• Command patterns: Which commands you run where
• Usage frequency: Weights suggestions by actual usage

The more you use SSH, the smarter the suggestions become!

• Local learning: All learning data stays on your machines
• No external calls: Hooks run entirely locally
• Private repo: Your configuration and usage patterns remain private

• Destructive operation warnings: rm -rf, chmod 777, etc.
• Remote sudo detection: Alerts for elevated remote operations
• Path validation: Catches common path mistakes

Add your own hooks to the hooks/ directory and they'll be synced across hosts.

The router automatically reads your ~/.ssh/config for available hosts.

• Check system status: ~/.claude/claude-sync/bootstrap.sh status
• View all statistics: ~/.claude/claude-sync/claude-sync-stats
• View SSH learning: ~/.claude/claude-sync/claude-sync-stats ssh
• View bash optimization: ~/.claude/claude-sync/claude-sync-stats bash
• Reset learning: rm ~/.claude/ssh_topology_*.pkl ~/.claude/learning_*.json

This is a private repo for personal use. Fork and customize for your own needs!

• Claude Code Documentation
• Claude Code Hooks Guide

🌟 Matrix-like superpowers for the command line!