Is this a background daemon process?

No. The /loop command handles exactly one task per invocation and returns control to the user, adhering to safety protocols while enabling high-velocity development.

How does the /loop command work?

The /loop command scans your local .agents/TASKS/ directory for the highest priority 'To Do' task, claims it by updating the metadata with a timestamp, and proceeds to implement the feature based on the linked PRD.

Can I use this with multiple AI editors simultaneously?

Yes, the skill uses a file-based locking mechanism (claims) that allows Claude CLI, Cursor, and Codex to work on different tasks in parallel without technical conflicts.

What happens if an agent is rate-limited?

When rate-limited, the agent saves its current progress to the task notes, releases the claim, and prompts you to switch to a different platform to continue the work where it left off.

AI Development Loop

Name: AI Development Loop
Author: shipshitdev

byshipshitdev

•

生産性とワークフロー

Orchestrates autonomous development tasks across multiple AI platforms with built-in QA gates and task management.

The AI Development Loop skill transforms your development workflow into an autonomous agent-driven engine by managing a task queue located within your repository. It allows AI agents to independently claim, implement, and move tasks through a structured lifecycle—from backlog to implementation and testing—while maintaining human-in-the-loop control through manual QA gates. By supporting parallel execution across Claude CLI, Cursor, and Codex, it maximizes productivity and effectively manages API rate limits, ensuring continuous progress on your feature backlog without manual hand-holding.

主な機能

01Autonomous task claiming and execution from a local Markdown-based queue

02Automated Git workflow integration with per-task branching and clear commit history

03Smart rate limit handling with seamless platform switching and state preservation

04Multi-platform orchestration supporting Claude CLI, Cursor, and Codex

0510 GitHub stars

06Human-in-the-loop QA gates to approve or reject code before completion

ユースケース

01Managing a high-volume feature backlog with autonomous implementation

02Enforcing standardized quality checks and human reviews in an AI-driven development cycle

03Distributing development tasks across different AI tools to bypass individual platform rate limits

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add shipshitdev/library ai-dev-loop

For use in Claude.ai and ChatGPT

Download Skill