Quick Start

Get feedback from AI agents on your CLI tools in under 2 minutes.

Install

# Option 1: Run directly with uvx (no install needed)
uvx --from git+https://github.com/chriskd/focusgroup focusgroup --help

# Option 2: Install as a tool
uv tool install git+https://github.com/chriskd/focusgroup

# Option 3: Clone and install for development
git clone https://github.com/chriskd/focusgroup && cd focusgroup
uv pip install -e ".[dev]"

Requires Python 3.11+ and uv.

Authenticate

Focusgroup uses CLI tools for agents. Set up the providers you want:

# For Claude
claude auth login

# For Codex (OpenAI)
codex auth

See providers for detailed setup.

Run

focusgroup ask "Is this help clear?" -x "mytool --help" -n 1

Output:

Agent-1 (claude):
The help output is mostly clear, but I have a few suggestions:

1. The "--format" flag mentions "json, text" but doesn't explain the default
2. The synopsis shows [OPTIONS] but doesn't indicate which are required
3. Consider adding example commands at the bottom

Overall: 7/10 - functional but could be more discoverable for new users.

Session saved: ~/.local/share/focusgroup/logs/20260106-abc123.json

That's it! You've consulted an AI agent about your tool's usability.

More Examples

# Ask 3 agents (default) for diverse perspectives
focusgroup ask "What improvements would help agents use this?" -x "mytool --help"

# Let agents actually run the tool (exploration mode)
focusgroup ask "Try common workflows" -x "mytool --help" --explore

# JSON output for piping
focusgroup ask "Review this" -x "mytool --help" -o json | jq .

# Context from a file
focusgroup ask "Review this API" -x "@README.md"

# Context from stdin
cat docs.md | focusgroup ask "What's missing?" -x -

# Dogfood: review focusgroup itself
focusgroup demo

Next Steps

• configuration - Config file reference for structured sessions
• modes - Session modes (single, discussion, structured)
• exploration - Letting agents run your tool
• providers - Agent provider setup and troubleshooting