tasak 0.1.3

Transform your AI coding assistant into a productivity powerhouse with custom tools and workflows

TASAK: The Agent's Swiss Army Knife

Transform your AI coding assistant into a productivity powerhouse with custom tools and workflows tailored to YOUR codebase.

📋 See what's new in v0.1.0 →

🚀 Why TASAK?

For AI Agent Power Users (Claude Code, Cursor, Copilot)

Problem: Your AI assistant wastes tokens rediscovering your project structure, can't run your custom toolchain, and you're copy-pasting commands back and forth.

Solution: TASAK gives your AI agent a curated toolkit that understands YOUR workflow:

• 📦 Package complex workflows into simple commands ("deploy staging" instead of 10 manual steps)
• 🧠 Reduce context usage by 80% through hierarchical command discovery
• 🔧 Self-improving: Let your agent write Python plugins to extend its own capabilities!
• 🎯 Project-aware: Different tools for different projects, automatically

For Development Teams

Problem: Every developer has their own way of running tests, deployments, and dev environments. Onboarding is painful.

Solution: TASAK standardizes your team's workflow into a unified command palette:

• 🏢 Company-wide tooling in global config, project-specific in local
• 📚 Self-documenting: Your AI agent can explain and execute any workflow
• 🔒 Secure by default: Only expose what you explicitly allow
• 🚄 Zero friction: Works with any language, any framework, any toolchain

💡 Real-World Magic

# Your AI agent can now do THIS with a single command:
tasak deploy_review_app
# Instead of:
# 1. Check git branch
# 2. Build Docker image
# 3. Push to registry
# 4. Update k8s manifests
# 5. Apply to cluster
# 6. Wait for rollout
# 7. Run smoke tests
# 8. Post PR comment with URL

🎯 Perfect For

✨ Claude Code / Cursor / Copilot / Gemini CLI / Codex CLI / Users

• Build a custom toolkit that makes your AI assistant 10x more effective
• Stop wasting time on repetitive commands - let your agent handle them
• Create project-specific "skills" your AI can use intelligently

👥 Development Teams

• Standardize workflows across your entire team
• Make complex operations accessible to junior developers
• Document-by-doing: your commands ARE the documentation

🔧 DevOps & Platform Engineers

• Expose safe, curated access to production tools
• Build guardrails around dangerous operations
• Create approval workflows for sensitive commands

🎨 Open Source Maintainers

• Give contributors a standard way to run your project
• Reduce "works on my machine" issues
• Make your project AI-assistant friendly

🌟 Killer Features

🧩 Python Plugins (NEW!)

Your AI agent can write its own tools! Just ask:

"Create a plugin that formats all Python files and runs tests"

The agent writes the Python function, TASAK automatically loads it. Mind = blown. 🤯

🎭 Three Modes of Power

cmd apps - Quick & dirty commands

format_code:
  type: cmd
  meta:
    command: "ruff format . && ruff check --fix"

mcp apps - Stateful AI-native services

database:
  type: mcp
  meta:
    command: "uvx mcp-server-sqlite --db ./app.db"

curated apps - Orchestrated workflows

full_deploy:
  type: curated
  commands:
    - test
    - build
    - deploy
    - notify_slack

🔄 Hierarchical Config

Global tools + project tools = perfect setup

~/.tasak/tasak.yaml       # Your personal toolkit
./project/tasak.yaml      # Project-specific tools
= Your AI has exactly what it needs

⚡ Quick Start

1. Install (30 seconds)

pipx install git+https://github.com/jacekjursza/TASAK.git

2. Create Your First Power Tool (1 minute)

cat > ~/.tasak/tasak.yaml << 'EOF'
header: "My AI Assistant Toolkit"

apps_config:
  enabled_apps:
    - dev
    - test
    - deploy

# One command to rule them all
dev:
  name: "Start Development"
  type: "cmd"
  meta:
    command: "docker-compose up -d && npm run dev"

test:
  name: "Run Tests"
  type: "cmd"
  meta:
    command: "npm test && npm run e2e"

deploy:
  name: "Deploy to Staging"
  type: "cmd"
  meta:
    command: "./scripts/deploy.sh staging"
EOF

3. Watch Your AI Agent Level Up

# Your AI can now:
tasak dev      # Start entire dev environment
tasak test     # Run full test suite
tasak deploy   # Deploy to staging
# No more copy-pasting commands!

🎓 Real Use Cases

Use Case 1: Supercharge Your Claude Code

# .tasak/tasak.yaml in your project
header: "NextJS + Supabase Project"

apps_config:
  enabled_apps:
    - setup_branch
    - check_types
    - preview

setup_branch:
  name: "Setup new feature branch"
  type: "cmd"
  meta:
    command: |
      git checkout -b $1 &&
      npm install &&
      npm run db:migrate &&
      npm run dev

check_types:
  name: "Full type check"
  type: "cmd"
  meta:
    command: "tsc --noEmit && eslint . --fix"

preview:
  name: "Deploy preview"
  type: "cmd"
  meta:
    command: "vercel --prod=false"

Now your Claude Code can:

• Create and setup feature branches
• Run comprehensive type checks
• Deploy preview environments ...all without you typing a single command!

Use Case 2: Team Workflow Standardization

# Company-wide ~/.tasak/tasak.yaml
header: "ACME Corp Standard Tools"

apps_config:
  enabled_apps:
    - vpn
    - staging_logs
    - prod_deploy

vpn:
  name: "Connect to VPN"
  type: "cmd"
  meta:
    command: "openvpn --config ~/.acme/vpn.conf"

staging_logs:
  name: "Stream staging logs"
  type: "cmd"
  meta:
    command: "kubectl logs -f -n staging --selector=app"

prod_deploy:
  name: "Production deployment"
  type: "curated"
  commands:
    - name: "deploy"
      description: "Full production deployment with approvals"
      backend:
        type: composite
        steps:
          - type: cmd
            command: ["./scripts/request-approval.sh"]
          - type: cmd
            command: ["./scripts/deploy-prod.sh"]

Use Case 3: Python Plugins - Let AI Extend Itself!

# Your AI agent can write this!
# ~/.tasak/plugins/my_tools.py

def smart_refactor(file_pattern: str, old_name: str, new_name: str):
    """Refactor variable/function names across multiple files"""
    import subprocess
    result = subprocess.run(
        ["rg", "-l", old_name, file_pattern],
        capture_output=True,
        text=True
    )
    files = result.stdout.strip().split("\n")

    for file in files:
        subprocess.run([
            "sed", "-i", f"s/{old_name}/{new_name}/g", file
        ])

    return f"Refactored {len(files)} files"

# Now available as: tasak smart_refactor "*.py" "oldFunc" "newFunc"

📚 Documentation

Quick Links:

• Why TASAK? - See more use cases and benefits
• Installation & Setup - Get running in 2 minutes
• Basic Usage - Your first cmd apps
• Advanced Usage - MCP servers, Python plugins, and workflows
• Changelog - See all releases and changes

🤖 CLI Semantics for Agents

For MCP and MCP‑Remote apps, TASAK presents a predictable, agent‑friendly CLI:

• tasak <app> → prints only tool names (one per line). No headers or descriptions.
• tasak <app> <tool> →
  • If the tool has no required parameters: executes immediately with empty args.
  • If the tool has required parameters: shows focused help for that tool (same as --help), including description and parameters with required/type info.
• tasak <app> <tool> --help → always shows focused help for that single tool.
• tasak <app> --help → prints grouped simplified help:
  • " commands:" — tools without required params (can run immediately) as <name> - <description>
  • " sub-apps (use --help to read more):" — tools with required params as <name> - <description>

Behavior notes:

• Tool schema listing/help uses a transparent 1‑day cache; when stale or missing, TASAK refreshes quietly and updates the cache.
• Noisy transport logs are suppressed by default; enable with TASAK_DEBUG=1 or TASAK_VERBOSE=1 if you need to debug.

Daemon (Connection Pooling)

TASAK can run a local daemon to pool MCP connections and cache schemas, dramatically reducing per-command startup time. The daemon runs on 127.0.0.1:8765 and the CLI auto-starts it on demand (unless explicitly stopped or disabled).

• Start: tasak daemon start
• Stop: tasak daemon stop (also disables autostart until next manual start)
• Restart: tasak daemon restart
• Status: tasak daemon status
• Logs: tasak daemon logs -f

Logging levels

By default the daemon is quiet (warning and errors only). Enable verbose logs when debugging:

• CLI flags:
  • tasak daemon start -v or tasak daemon restart -v (equivalent to debug)
  • tasak daemon start --log-level info (or debug, warning, error)
• Environment variable:
  • TASAK_DAEMON_LOG_LEVEL=INFO (or DEBUG) before starting the daemon

CLI-side daemon hints ("Using daemon…", "Daemon: …") appear only when --debug or TASAK_VERBOSE=1 is set.

HTTP endpoints

The daemon exposes a small local API for health checks and diagnostics:

• GET /health – basic liveness + uptime
• GET /connections – active connections with age/idle and counters
• GET /apps/{app}/ping?deep=true – shallow or deep ping (deep performs a quick tool list)
• GET /metrics – basic counters (connection creations/reuses, per-app list/call/error counts)

Autostart behavior

The CLI auto-starts the daemon unless one of the following is true:

• tasak daemon stop was called (creates ~/.tasak/daemon.disabled)
• TASAK_NO_DAEMON=1 is set in the environment
• --debug is used (bypasses daemon for direct connections)

Tuning

You can tune TTLs via environment variables before starting the daemon:

• TASAK_DAEMON_CONN_TTL – connection idle TTL in seconds (default: 300)
• TASAK_DAEMON_CACHE_TTL – tools cache TTL in seconds (default: 900)

🤝 Community & Support

• GitHub Issues: Report bugs or request features
• Discussions: Share your TASAK configs and workflows
• Examples: Check out examples/ folder for real-world configurations

🛠️ For Contributors

Built with Python 3.11+, following TDD principles. We welcome contributions!

Development Setup

git clone https://github.com/jacekjursza/TASAK.git
cd TASAK
python -m venv .venv
source .venv/bin/activate

# Install in editable mode (includes MCP by default)
pip install -e .

# Run tests
pytest -q

# Optional: if you install pytest-timeout, you can enable
# suite timeouts using the provided CI config
pytest -c pytest-ci.ini -q

See CONTRIBUTING.md for guidelines.

📄 License

MIT License - see LICENSE for details.