Council is a skill that coordinates multiple agent CLIs on your machine to deliberate across turns before presenting a final plan. It operates locally via one-shot CLI invocations and keeps transcripts for audit.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Council is an agent skill that coordinates multiple agent CLIs installed on your machine to deliberate for a specified number of turns before producing a final plan. It drafts a plan and presents it independently to each agent CLI, collects their responses, revises the plan, and outputs the full deliberation plus the final plan. All invocations are read-only with no external APIs.
How it works
- The lead agent drafts a plan.
- Each registered agent CLI on the machine is invoked once per turn and returns a verdict: SHIP IT / SHIP WITH CHANGES / RETHINK.
- Members’ replies are not shared with other members.
- The plan is revised with member feedback and repeated for the requested number of turns.
- The final output includes the entire deliberation transcript and the final plan.
Getting started
Install by cloning the SKILL.md into the appropriate skills directory for your agent:
# Claude Code (personal)
git clone https://github.com/kitze/council ~/.claude/skills/council
# Claude Code (per-project)
git clone https://github.com/kitze/council .claude/skills/council
For other agents, clone into their skills directory or provide the SKILL.md as instructions.
Requirements
At least one other agent CLI installed and logged in, such as: claude, codex, cursor-agent, gemini, grok, or opencode. The more you have, the bigger the council.
Notes
Deliberation sessions and transcripts are stored in ~/.council/ for auditing. The SKILL.md embeds headless flags for each CLI (e.g., cursor --trust, gemini --skip-trust, codex read-only sandbox, grok --system-prompt-override) to facilitate operation.
License
MIT