Show HN: Todo-CLI – Add a todo, AI agents complete it for you (github.com)

Todo-CLI is a terminal-based task manager that spawns persistent AI “personal assistant” agents to finish your todos for you. Built around Gemini CLI and MCP (micro‑connector) tools, each task gets its own tmux session running a Gemini agent; you get a blessed-based full-screen TUI with real-time split‑screen previews, Vim-like navigation, external-editor support, and persistent JSON storage (~/.config/todo-cli/tasks.json via lowdb). Agents can be prompted to check Gmail, make phone calls (ElevenLabs), search services, schedule meetings (Google Calendar), research and compile reports, and more. Key features include YOLO mode (auto-approve agent actions), agent output captured every second, session isolation and auto‑kill on delete, and easy system-prompt customization via .gemini/system.md. Code is MIT-licensed and packaged as an npm app. This is significant because it moves a todo list from a tracker to an executor, showing a practical pattern for orchestrating LLM-driven agent workflows locally. Important technical and security implications: MCP servers run with your user permissions (review their code and trust sources), personal context is provided via plain text files in /user_data (simple persistent memory), and there is currently no automated hook to mark tasks done—status updates remain manual. Setup for MCPs is still clunky and better personal-info management, multi-agent selection (other LMs), and improved extension installation are suggested next steps.