gardenvision

AI Coordinator Pattern — AI skill for Claude Code

AI community

One human, many AI sessions, no chaos: a coordinator pattern for Claude Code, Codex and other AI tools (EN/DE, CC BY 4.0).

How to install AI Coordinator Pattern

This entry records only its repository, not the path inside it, so there is no exact command to give. Open gardenvision/ai-coordinator-pattern and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What AI Coordinator Pattern does

One human, many AI sessions, no chaos: a coordinator pattern for Claude Code, Codex and other AI tools (EN/DE, CC BY 4.0).

Alternatives in AI

Omnara — by Ishaan Sehgal - A command center for AI agents that syncs Claude Code sessions across terminal, web, and mo 2.6k ★
Real-World Pain Points — Project: https://github.com/Narwhal-Lab/MagicSkills For many teams building multi-agent systems, the first thi 277 ★
Laravel AI Translator — High quality automatic translate your language files into many languages using AI like Claude, GPT and etc 250 ★

README

AI Coordinator Pattern – Summary

![AI Coordinator](images/ai-coordinator.jpg)

**English** · [Deutsch](README.de.md)

Version 1 · As of 2026-10-21 · Summary of [AI-Coordinator-Pattern.md](en/AI-Coordinator-Pattern.md) (version 4.4) · for sharing

**What it is about:** One human lets many AI sessions work on one project at the same time. One of them is the **coordinator**. It hands out the work, assigns scarce resources, has results checked independently, keeps a log and puts decisions in front of the human. The human decides; the AI does the work.

The pattern grew out of practice: two computers, more than 20 parallel sessions, several AI providers. It fits any project with many AI sessions – software, data, research or documentation. It works with Claude Code, Codex, OpenCode and similar tools.

The problem

As soon as several AI sessions run in parallel, the same things happen again and again:

Two sessions work in the same program at the same time and break each other's work.
One session says "the other one already agreed" – but the human was never asked.
Results exist only in the chat. One message gets lost, and everyone waits.
Two sessions research the same thing, and the quota runs out faster than needed.
Nobody knows any more which state is the current one.

The idea in one picture

                 Human (decides, gives approvals, tests)
                                    │
                        Coordinator (exactly one)
      ┌─────────────┬───────────────┼────────────────┬──────────────────┐
 Task session   Task session     Helper          Station lead       External worker
 (per task)     (per task)     (short job,       (per extra         (other model:
                                own copy)         computer)          review, research)

The coordinator builds little itself. It keeps the overview, sets the order of work and checks the results.