kazunori279

Issuelooper — AI skill for Claude Code

AI community

A swarm development framework, your personal software factory: AI coding agents (Antigravity CLI or Claude Code) build your repository in parallel, coordinated only through GitHub issues, labels and p.

How to install Issuelooper

This entry records only its repository, not the path inside it, so there is no exact command to give. Open kazunori279/issuelooper and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Issuelooper does

Alternatives in AI

Orkas — Open-source multi-agent AI desktop client — build and command your AI agent team through conversation 1.5k ★
Claude Hub — by Claude Did This - A webhook service that connects Claude Code to GitHub repositories, enabling AI-powered c 393 ★
Ralph Loop — A long-running AI agent loop 292 ★

README

issuelooper

A swarm development framework: your personal software factory. You file issues, answer the occasional question and review what ships; a team of AI coding agents does the rest.

issuelooper develops a repository with many AI coding-agent sessions at once, coordinated only through GitHub. Issues are the work queue, labels are the state, pull requests are the output, and signed comments are the conversation. A deterministic supervisor (a program, not an agent) watches the repository, starts a fresh agent session for each unit of work, and merges pull requests that pass its gates.

It supports the Antigravity CLI (`agy`) and Claude Code (`claude`) as the agent runtime, per role.

How it works

             owner files issues
                     │
                     ▼
   ┌──────────── monitor (one long-lived session) ────────────┐
   │  triage · priorities · holds · dependencies · decisions  │
   └──────────────────────────────────────────────────────────┘
                     │ stage:triage
                     ▼
 supervisor (ccpool.py, polls GitHub every 15 s, one GraphQL call)
   ├─ worker   one per claimable issue: failing test → fix → sweep → PR
   ├─ merge    no agent: review gate + merge gates → squash-merge → deploy
   ├─ merger   one per conflicting PR the supervisor could not rebase
   └─ tester   live checks after merges; UX runs and spec checks of the app

Role	Kind	Does
Supervisor	Program (`ccpool.py`)	Turns GitHub state into units of work, starts them, merges, deploys.
Worker	Agent session, one per issue	One issue: a failing test, the fix, a green sweep, a pull request.
Merger	Agent session, one per conflicting PR	Rebases, resolves so both sides survive, sweeps again, merges.
Tester	Agent session (optional)	Live checks against the real environment; UX runs that walk the deployed app and file what gets in a user's way; spec checks of the app against the spec, one