How to use Claude to extend a reviewed task runner
Claude Code extended Ralph through its own queued plans, adding worktrees and review loops while Stevens retained architecture choices and blocked-task stops.
linkedin.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Extend this task runner from a single repository to multiple repositories using the approved design. Preserve fresh sessions per work item, durable task state, isolated worktrees and pull-request review feedback. Stop with a written reason when blocked, and distinguish completed behavior from queued features before requesting acceptance.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Gethin Stevens
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The owner reports a working runner that extended itself from a single-repository skeleton into a reviewed multi-repository task system over roughly three days.
- Human involvement
- Stevens made architectural choices, specified work and unblocked failures; pull-request review remained part of the implementation loop.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Stevens supplied design plans and Claude Code implemented a minimal Ralph runner.
- He fed the runner its own feature plans and bugs, adding queue, worktree, pull-request and review-feedback behavior.
- A human-directed architecture change expanded the runner across repositories, while stalled tasks stopped with a durable explanation.
Gethin Stevens first wrote design plans for Ralph, a task runner around Claude Code. Claude implemented a minimal working skeleton. Stevens then fed that runner plans and bugs for its own development, so the queue became a way to extend the system that operated it.
The resulting work added queue management, Git worktrees, pull-request handling and a review-feedback loop. A fresh agent session handled a work item, while durable files preserved the task state. If progress could not be made honestly, the agent wrote STUCK.md and stopped. A cloud review bot examined pull requests, and another local loop read feedback and applied corrections.
A later human plan expanded the original single-repository design across multiple repositories. Stevens reports roughly three days of steering and architectural decisions without writing implementation code himself. He still unblocked stalled work and chose what the system should become.
The article links the public repository and acknowledges that its README lagged behind the running architecture. Automatic issue creation from failing builds or log spikes was still being built when the article was written, so it is not counted as a finished result. The concrete outcome is a runner extended through its own reviewed backlog, with explicit stops and human direction.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: LinkedIn article · Gethin Stevens
Frequently asked questions
Did Ralph decide its own product goals?
No. Stevens supplied plans, architectural decisions and interventions when work stalled.
What happened when an agent could not progress?
It wrote STUCK.md with the reason and stopped the task.
Was automatic log-to-issue creation already complete?
No. That feature was still in Ralph’s queue when the article was written.
More details
About this account
Stevens supplied design plans and Claude Code implemented a minimal Ralph runner. He fed the runner its own feature plans and bugs, adding queue, worktree, pull-request and review-feedback behavior.
Firsthand dated account opened. Claude Code implementation is explicit. Reported elapsed time and no-handwritten-code claim are attributed to the owner; unfinished automatic issue ingestion is excluded. Suggested prompt written by Agents In Use; no original task prompt was found in the verified text.
Credit: Gethin Stevens — LinkedIn article
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude Code · Ralph · Git worktrees · GitHub pull requests
self-hosted development · durable task state · review feedback
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.