Analysis updated 2026-05-18
Drop the .claude folder into a project and give Claude Code a request, it will break the work into GitHub issues, dispatch subagents to implement each one, and manage the PR review loop.
Use the Stop hook to enforce that every subagent session produces a pull request before it ends, preventing silent incomplete work.
Pin subagent model versions in the agent definition files to control cost when the main orchestrator session runs on an expensive model.
| nickalaslight/fable-sonnet-orchestrator-kit | blackvenom5iix/winget-toctou-poc | grafikerdem/agent-context-economy | |
|---|---|---|---|
| Stars | 1 | 0 | 6 |
| Language | PowerShell | PowerShell | PowerShell |
| Setup difficulty | moderate | moderate | easy |
| Complexity | 3/5 | 3/5 | 2/5 |
| Audience | developer | researcher | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires Claude Code, the GitHub CLI authenticated non-interactively, and PowerShell 7+ installed, three placeholder values must be replaced before use.
This repository is a set of configuration files for Claude Code, an AI coding assistant, that wires up a specific way of working on GitHub projects. The goal is to split the AI's role in two: one main agent acts as a manager that breaks down requests into GitHub issues and hands each issue off to a separate focused subagent to implement. The subagents work on isolated copies of the code, open pull requests, and the main agent reviews and merges them. The workflow goes in a fixed loop: a request comes in, the main agent turns it into one or more GitHub issues, a subagent per issue creates a separate working branch, makes the smallest correct change needed to close that issue, opens a pull request, and the main agent reviews it. A PowerShell script enforces the hand-offs at every step. If a subagent finishes its turn without opening a pull request, the script blocks it from proceeding until one exists. If the main agent has requested changes on a pull request, the script injects the review feedback into the subagent's context so it fixes the issues in that same session. The practical benefit is that this structure keeps the two agents from doing each other's jobs. The main agent handles judgment and sequencing. The subagents handle implementation. The configuration pins the subagents to a specific Claude model version so that cost does not accidentally escalate if the main session is running on a more expensive model. A rolling handoff document keeps a summary of current state so that if the main agent is interrupted or restarted, a fresh instance can pick up where the previous one left off. Installation is copying the .claude folder into your project and replacing three placeholder values with your GitHub organization, local checkout path, and secrets directory path. No credentials are stored in the repository files themselves.
A Claude Code configuration kit that turns the main AI agent into an orchestrator: it creates GitHub issues, dispatches focused subagents per issue to a git worktree, and a code-enforced hook gates PR creation and review before merging.
Mainly PowerShell. The stack also includes PowerShell, Claude Code, GitHub CLI.
No license file is included in the repository.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.