Show HN: Todo-CLI – Add a todo, AI agents complete it for you

6 hours ago 1

The todo app that actually gets things done.

A terminal-based task manager where AI agents finish your todos for you. Powered by Gemini CLI and MCP tools. Why? The best todo list isn't one that tracks tasks—it's one that completes them. Inspired by claude-squad and Conductor, todo-cli runs multiple AI agents in parallel, each working on a different task. But instead of coding agents, these are personal assistant agents that can:

📧 Check your email and make phone calls
🗺️ Search for services and request quotes
📅 Schedule meetings and manage your calendar
🔍 Research topics and compile reports ...and anything else you can describe in a prompt

📋 Terminal UI - Full-screen todo list built with blessed
🤖 AI Agent Integration - Each task automatically gets a Gemini CLI agent in a tmux session
👀 Real-time Preview - Watch agent output in split-screen view (40/60 layout)
⌨️ Keyboard-driven - Vim-like navigation and shortcuts
💾 Persistent Storage - Tasks saved to ~/.config/todo-cli/tasks.json
🎯 YOLO Mode - Optional auto-approval for agent actions
✏️ External Editor - Press Ctrl+E to edit in vim/nano

Node.js v18+
tmux - Installation Guide
Gemini CLI

Allow Gemini-cli to use custom system prompt under .gemini

There is an example system.md under .gemini, it includes ElevenLabs Agent prompting guide to optimize phone calls. Update the system prompt to adjust what you want to agent to do

Setup MCP or Extention to extend agent capability

Some useful MCPs:

Google Map to get location data
Gmail to access gmail, get information or send out email
Google Calendar to access calendar, get events or update calendar
Eleven Labs to create conversational agents for phone calls

Note: MCP servers run with the same permissions as your user account. Only connect to trusted MCP servers and review their code before use.

Provide Context for Your Agent

Create text files in the /user_data directory to give your agent access to important information. The agent can read these files when needed using Gemini CLI's built-in file reading and search capabilities.

This simple file-based approach provides persistent memory - your preferences, account details, and context are stored as regular text files that you can easily view and edit.

Start the application:

a - Add new task (title → details → YOLO mode)
Enter - Open/attach to agent session (Ctrl+q to return)
x - Toggle task done/undone
d - Delete task and kill session
j/k or ↑/↓ - Navigate tasks

[ - Scroll preview up
] - Scroll preview down
g - Jump to bottom of preview

q - Quit application

Enter - Submit and move to next step
Ctrl+E - Open external editor ($EDITOR) for full cursor control
Escape - Cancel

When you add a task, you'll be prompted for:

Task Title - Brief description of what to do
Details (optional) - Additional context for the agent
YOLO Mode (y/N) - Auto-approve all agent actions

Tip: Press Ctrl+E in any input field to open your $EDITOR (vim/nano) for full text editing with cursor movement!

┌─ TODO ──────────────┬─ AGENT PREVIEW ──────────────────┐ │ │ │ │ > ☐ Task 1 │ [Session: gemini-task-abc123] │ │ ☐ Task 2 │ [Started: 10:30:45 AM] │ │ ✓ Task 3 (done) │ │ │ │ Agent output appears here... │ │ │ │ │ [3 tasks] │ │ │ │ │ └─────────────────────┴───────────────────────────────────┘ Status: ✓ Task created: gemini-task-abc123 [a]Add [Enter]Open session(Ctrl+q) [x]Done [d]Delete | "["/""]" Scroll [g]Bottom [q]Quit

blessed - Terminal UI framework
lowdb - JSON file storage
tmux - Process isolation for agent sessions
nanoid - Unique ID generation

todo-cli/ ├── bin/ │ └── todo-cli.js # Entry point with dependency checks ├── src/ │ ├── app.js # Main application logic │ ├── models/ │ │ └── task.js # Task data model │ ├── services/ │ │ ├── storage.js # lowdb persistence layer │ │ └── tmux.js # tmux session management │ └── ui/ │ ├── screen.js # blessed UI layout │ └── modal.js # Task creation modal └── package.json

Create Task - Enter title, details, and YOLO mode preference
Spawn Agent - Creates a tmux session running gemini -i "{prompt}" (or with --yolo)
Preview Output - Captures tmux pane output every second
Auto-scroll - Shows latest agent output automatically
Manual Control - Scroll up with [ to read history, press g to jump to bottom
Attach Session - Press Enter to attach full-screen, Ctrl+q to return

Cursor Movement: Press Ctrl+E in input fields to open vim/nano for full editing
YOLO Mode: Tasks with YOLO mode enabled let agents execute actions without confirmation
Auto-scroll: Preview auto-scrolls to show new output; scroll up manually to pause
Session Isolation: Each task runs in an isolated tmux session that persists until deleted
Clean Termination: When you delete a task, its tmux session is automatically killed

Come up with a better approach to manage personal info as context.
Allow using different agent (codex, claude-code) if they have a way to replace the system prompt
Communicate between agent and todo tasks, right now gemini-cli doesn't have hooks, so there's no direct way check if the task is done or not. Still needs manual update
Better MCP/Extension installation, right now its still cumbersome to setup

MIT

Read Entire Article