TaskCLI — General-Purpose AI Coding Agent

⚠️ IMPORTANT SAFETY WARNING ⚠️ This tool is EXPERIMENTAL and executes commands autonomously with AI. It can:

• Modify, create, and delete files in your filesystem
• Execute shell commands with real effects
• Make irreversible changes to your project

USE AT YOUR OWN RISK. Always:

• Work in a backed-up or version-controlled environment
• Review actions before confirming (use without -y flag)
• Test in a safe directory first
• Never run with sudo or elevated privileges

This is a "YOLO" tool - it takes actions first and asks questions later. You have been warned! 🚨

Overview

TaskCLI is a general-purpose AI coding agent that can work with ANY programming language or framework:

• Universal: Works with Python, JavaScript, TypeScript, Go, Rust, Java, C++, and more
• Framework Agnostic: React, Vue, Django, Rails, Spring, Express - it handles them all
• Full-Stack: From frontend UI to backend APIs to DevOps scripts
• Autonomous: Breaks down complex goals and executes them step-by-step
• Self-Correcting: Runs tests, catches errors, and fixes them automatically

Setup

1. In TaskCLI/, copy .env.example to .env and set GEMINI_API_KEY (or GOOGLE_API_KEY)

Install

cd TaskCLI
npm install

Quick Health Check

npm run doctor

Usage

# Interactive UI (default)
node bin/taskcli.js

# Headless (no UI)
node bin/taskcli.js --headless "Build an Express API skeleton with tests"

# Auto-confirm shell commands
node bin/taskcli.js -y --headless "Initialize a Node project with chalk and execa"

# Use legacy v1 orchestrator (uses Gemini Flash for planning)
node bin/taskcli.js --v1 --headless "Create a TypeScript CLI"

Non-Interactive Smoke Test (no network)

# Runs a canned plan that verifies shell, file write/read, codegen and execution
npm run smoke

Context Persistence

• TaskCLI creates a .taskcli folder in your working directory to save session context
• On startup, it detects previous sessions and offers to resume
• Context is automatically saved after each interaction
• Use slash commands in interactive mode:
  • /resume - Resume from saved context
  • /save - Manually save current context
  • /status - Show context status
  • /clear - Clear all context and start fresh
  • /init - Inspect project structure (read-only)
  • /help - Show all available commands

Interactive Features

• Smart slash command palette with filtering and tooltips
• Status bar showing current working directory
• Command history with arrow key navigation
• Context-aware session management

Notes

• Task planning and execution uses Google's GenAI SDK directly with Gemini Pro
• The system prompt is defined in src/agent.js as AGENT_SYSTEM_PROMPT. You can edit this constant to customize AI behavior.
• Pro calls include full session context (prior goals, commands, outputs) to maintain continuity across tasks.
• Web search uses googlethis to supplement context
• Interactive UI is built with Ink (React for CLIs). If your terminal doesn't support raw mode, the CLI falls back to non-interactive.