AI Agent Workflow Console

A local terminal-based multi-agent interface (TUI) that orchestrates AI workflows using Claude CLI.

Concept

A developer productivity console where each AI agent performs a specific role in a 4-stage workflow:

PROMPT → PLAN → EXECUTE → REVIEW
              ↑_______________|
                (orchestrated)

Features

• 4-panel TUI — Prompt editor, Plan output, Execute output, Review panel
• AI-driven orchestrator — Claude decides which agent runs next
• Structured agent contracts — Each agent produces deterministic, parseable output
• Iterative review cycles — REVIEW can send work back to PLAN or EXECUTE
• Shared workspace context — All agents know the project state
• Session persistence — SQLite-backed sessions with resume capability
• Streaming output — Real-time token-by-token display in panels

Stack

• Python + Textual — TUI framework
• asyncio subprocess — Claude CLI integration
• Pydantic — Output contract validation
• aiosqlite — Session persistence
• uv — Package manager

Keyboard Shortcuts

Shortcut	Action
Ctrl+Enter	Send prompt to PLAN agent
Ctrl+P	Regenerate plan
Ctrl+E	Execute current plan
Ctrl+R	Trigger review
Tab	Switch panels

Requirements

• Python 3.11+
• uv
• Claude CLI with claude --dangerously-skip-permissions

Getting Started

# Install dependencies
uv sync

# Run
uv run python -m app

Project Structure

app/
├── agents/          # Agent prompts and output contracts
├── core/            # Subprocess runner, output parser, persistence
├── orchestrator/    # FSM and AI-driven routing
└── ui/              # Textual panels and layout

---

Work in progress — see .planning/ROADMAP.md for current status