Hand tickets to coding agents that open the pull requests for you
Idea + starter prompt, not run by us yet
For: Engineering teams with a ticket backlog and the skills to run their own Kubernetes cluster
Optio is a self-hosted tool that runs on Kubernetes. You give it a ticket from GitHub, Linear or Jira. It prepares a pod for the repo, runs Claude Code or GPT's Codex CLI there and opens a pull request. A watcher follows CI and review, and sends the agent back in when checks fail or reviewers comment. The README says merging follows green CI.
Steps
- Check the prerequisites first: Postgres, Redis, Helm and a Kubernetes cluster at version 1.33 or higher.
- Install it on a test cluster and a throwaway repo, not your main project.
- Start with one small, low-risk ticket and read the pull request line by line.
- Set the limit on automatic resumes so a stuck pull request stops and waits for you.
- Keep human review required before merge until you trust the results. The README gives no success rates.
- Watch the cost record it keeps for each task.
Starter prompt
I am considering Optio, a self-hosted tool that turns tickets into pull requests using coding agents on Kubernetes. Read its README at [repo link] and my cluster notes: [Kubernetes version, whether Helm, Postgres and Redis are available]. This job only reads and plans. Install nothing, create no token and run no cluster command until I say yes. Plan a trial first: a test cluster, a throwaway repo and one dummy low-risk ticket, such as fixing a typo in a README. Nothing in the trial touches my real repos or real tickets. Then give me the plan: what I must prepare, which settings to limit (including how many automatic resumes are allowed), and how to keep a person reviewing before merge. List the risks: cost, agents touching the wrong files, and pull requests that get stuck. Stop and wait for my yes before each install, helm or kubectl command, token creation, pull request, merge or deploy. Showing me a command is not permission to run it. Do not ask me for tokens in this chat. Tell me which secrets I should set up myself.
Original post: GitHub · found via agentusecases.ai. Written up in our own words. We haven't run it, so test it on made-up data first.