How does the skill handle context resets or crashes?

It implements a re-anchoring protocol that reads the current state from a local file and verifies git history at the start of every iteration, allowing it to resume work accurately.

What happens if the AI gets stuck in a loop?

The skill includes safety rails such as a global iteration limit (default 50) and specific phase iteration caps to prevent the agent from getting stuck indefinitely.

What makes this different from a standard AI prompt?

Unlike standard prompts, this skill uses a structured 'Phase Machine' and a persistent JSON state file to manage long-running tasks across multiple context windows and iterations.

Can I use this alongside the Ralph Loop?

No, because both utilize stop hooks, they will conflict. You must cancel the Ralph Loop using the /cancel-ralph command before starting the Autonomous Loop.

Autonomous Engineering Loop

Name: Autonomous Engineering Loop
Author: dha201

bydha201

0•

生産性とワークフロー

Automates long-horizon engineering tasks through a multi-phase workflow with built-in state persistence and crash recovery.

The Autonomous Engineering Loop is a robust framework for Claude Code designed to manage complex, multi-stage development projects without manual intervention. By utilizing a persistent JSON-based state machine, it guides Claude through distinct phases—from initial brainstorming and planning to implementation, testing, and shipping. This skill solves the problem of context drift during long sessions by implementing a re-anchoring protocol at every iteration, ensuring that even if the context resets, the autonomous agent knows exactly where it left off and what needs to happen next.

主な機能

01Automated re-anchoring protocol to prevent context drift between iterations

02Persistent state management via .claude/autonomous-loop.json for crash recovery

030 GitHub stars

04Seamless integration with compound engineering workflows for end-to-end delivery

05Multi-phase workflow covering brainstorming, planning, coding, and testing

06Built-in safety rails including iteration limits and fix-cycle caps

ユースケース

01Managing large-scale refactors that require multi-step validation and iteration

02Automating the complete PR lifecycle from initial concept to final shipping

03Executing complex, multi-day feature developments autonomously

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add dha201/architecture-documentation-plugin autonomous-loop

For use in Claude.ai and ChatGPT

Download Skill