fynnfluegge/agtx

What it solves

agtx is a terminal-native agentic development environment designed to move beyond the "one agent, one task" limitation. It solves the problem of shipping complex software where tasks are too interdependent to split cleanly, allowing multiple different coding agents to collaborate autonomously on a shared "blackboard" (a kanban board with a dependency graph).

How it works

It implements a blackboard system architecture where a shared knowledge base (the kanban board) is updated by various specialist coding agents (knowledge sources).

• Isolation: Each agent runs in its own dedicated git worktree and tmux window, ensuring they don't interfere with each other's context.
• Orchestration: A control shell manages the task lifecycle (Backlog $\rightarrow$ Planning $\rightarrow$ Running $\rightarrow$ Review $\rightarrow$ Done). It uses plugin phase gates and a dependency graph to determine when tasks can advance or start.
• Agent Switching: Users can configure different agents for different phases (e.g., Gemini for research, Claude for implementation).
• Integration: It provides an MCP server to allow agents to use skills like /agtx:brainstorm and /agtx:sweep to capture ideas and push them back to the board as tasks.

Who it’s for

Developers who use AI coding agents and want a structured, multi-agent workflow to manage complex projects without leaving the terminal.

Highlights

• Multi-Agent Collaboration: Support for a wide array of agents including Claude Code, Codex, Grok, Cursor, and Gemini CLI.
• Parallel Execution: Every task operates in its own git worktree and tmux window for simultaneous progress.
• Spec-Driven Plugins: Integration with frameworks like GSD, Spec-kit, and OpenSpec to define custom workflows.
• Brainstorm & Sweep: Specialized skills to explore ideas freely and then decompose them into actionable tasks on the board.
• Experimental Orchestrator: An AI agent that can autonomously manage the kanban board via MCP.