council-cli 0.1.2

Two AI agents brainstorm, debate, and research together

council-cli

Two AI agents brainstorm, debate, and research together using Claude Code.

Council launches two Claude Code instances that analyze your codebase in alternating rounds, tagging their points with AGREE/DISAGREE/REFINE to drive toward consensus.

Installation

pip install council-cli

Or install from source:

git clone https://github.com/david/council-cli
cd council-cli
pip install -e .

Prerequisites

• Claude Code CLI installed and authenticated (npm install -g @anthropic-ai/claude-code)

Usage

Standard Mode (formal rounds)

# 5-round council on error handling in current directory
council "Review error handling"

# 3 rounds on auth security
council -r 3 "Is the auth system secure?"

# Analyze a specific project, save to custom file
council -d ./my-project -o report.md "Code quality review"

# Verbose mode — see prompts being sent
council -v "Performance bottlenecks"

# Dry run — validate arguments without calling Claude
council --dry-run "Test topic"

Live Brainstorm Mode (--live)

Short, conversational back-and-forth exchanges instead of formal rounds. Feels like two engineers at a whiteboard.

# Default 10 exchanges
council --live "Should we refactor the auth system?"

# 20 exchanges for a deeper brainstorm
council --live --exchanges 20 "Performance vs readability"

Research Mode (--research)

Agents do deep web research and lateral thinking before brainstorming grounded in real data.

# Research mode with default 10 brainstorm exchanges
council --research "AI code review tools market"

# Research mode with 6 brainstorm exchanges
council --research --exchanges 6 "Developer productivity tools"

Flags

Flag	Description
topic	The analysis topic or question (required)
-r, --rounds N	Number of analysis rounds in standard mode (default: 5)
-d, --dir DIR	Working directory for analysis (default: .)
-o, --output FILE	Output file path for transcript (default: council_transcript.md)
-v, --verbose	Enable verbose output
--live	Live brainstorm mode
--research	Research mode (web search + lateral thinking + brainstorm)
--exchanges N	Number of exchanges in --live/--research mode (default: 10)
--dry-run	Validate arguments without running

How It Works

1. Standard Mode: Agent-Alpha analyzes the topic, Agent-Beta critiques, they go back and forth for N rounds, then produce a consensus.
2. Live Mode: Agents exchange short messages like a real conversation, then summarize key takeaways.
3. Research Mode: Alpha does deep web research, Beta does lateral thinking on the findings, then both brainstorm grounded in real data.

Output

Council saves a markdown transcript to council_transcript.md (or your -o path) with metadata, round-by-round discussion, and final conclusions.

License

MIT