The Council

Query multiple AI CLI coding assistants in parallel and compare their responses.

Default Tools

• Claude (Claude Code CLI)
• Gemini (Gemini CLI with gemini-3-pro-preview)
• Codex (OpenAI Codex CLI)
• Kimi Code (Kimi K3 via kimi)
• Cursor Agent (optional backend via --use-cursor)

Install

make install

Usage

# Default: read-only mode (can run gh/bash, no file writes)
council "Review this code for bugs"

# With project context
council -C /path/to/project "Analyze the architecture"

# Permission modes
council --locked "Explain this code"        # No shell, read-only
council --read-only "Review this PR"        # Default: shell OK, no writes
council --yolo "Fix all the tests"          # Unrestricted

# Use cursor-agent as backend for all tools
council --use-cursor "Compare error handling approaches"

# Pick specific tools
council -t claude,gemini,kimicode "Should I use Redis or Memcached?"

# Health check
council doctor

Permission Modes

Mode	Flag	Shell	File Writes
locked	--locked	No	No
read-only	--read-only (default)	Yes*	No
yolo	--yolo	Yes	Yes

* Kimi Code maps both locked and read-only to a bundled restricted agent profile because Kimi's --prompt mode auto-approves tools and cannot be combined with --plan. The profile does not expose shell, file-write, or subagent tools. This conservative mapping guarantees no workspace writes; Kimi can inspect with read-only tools but cannot run shell commands in these modes. yolo uses Kimi's plain --prompt mode, whose unattended AFK behavior auto-approves the full toolset.

Requirements

At least one of: claude, gemini, codex, kimi, agent (cursor-agent)