Use OpenClaw to run coding agents on PRs
A solo developer used OpenClaw to spin up Codex and Claude Code for coding, testing, commits, PRs, and review replies, while getting Telegram updates when work finished.
- What they did
- The developer set up OpenClaw, which they call Zoe, as the coordinator for coding work. It kept project context in Obsidian meeting notes, read-only production data, and an admin API, then launched Codex or Claude Code in a fresh git worktree and tmux session for each task with a prompt it generated. A cron check every 10 minutes watched for live sessions, open PRs, and passing CI, and if an agent failed, it retried up to 3 times with a rewritten prompt.
- How it went
- Over four weeks, they say the system reached as many as 94 commits in a day and averaged 50. They also reported 7 PRs finished in 30 minutes, with human review taking 5-10 minutes per PR. The main bottleneck was memory: a 16GB Mac Mini could handle only 4-5 agents at once.
- Tips
- Treat each PR as unfinished until CI and three AI reviewers pass, and plan for screenshots on UI work. Expect monthly costs of about $190 for Claude and Codex, with about $20 needed to get started.
Try it yourself with OpenClaw
Act as my coding-work dispatcher for [repo path or GitHub URL]. Read the task list in [issue tracker, project board, or file], pick one item, and assign it to either Claude Code or Codex. Create a fresh worktree and run [test command] after making changes, then commit the work and prepare a pull request. Before any irreversible step like opening the PR, ask me to approve the title and short diff summary. If a review comment comes in, draft a reply for me to approve before posting it. Send a message to me on [Telegram or other channel] when a PR is ready, tests fail, or a run needs attention. Do not merge, submit, or delete anything without my explicit approval.
Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.
Get 3 agent uses like this every morning
With the day's top AI news. Free, and one click to unsubscribe.
Related agent uses
Use Claude Code with Ollama for local coding help
They configured Claude Code to talk to a local Ollama model and used it on a small Java project inside IntelliJ.
Use Claude Code from a phone with a remote VM
They accessed Claude Code on an iPhone by pairing a Tailscale-only Vultr VM with Termius, mosh, and tmux.
Run Claude Code in a sealed VM for PRs
They confined Claude Code to a restricted virtual machine so it could draft changes and open PRs without direct access to secrets or merges.
Let Codex implement while you handle planning
One person used ChatGPT for planning and Codex for execution, with a small port helper to keep parallel local runs from colliding.
Run work coding and support tasks through agents
A consultant uses Claude, Codex, and a custom containerized harness to handle client software work and a personal assistant project.
Try a local search layer for code lookup
ColGrep-Agent-2B runs locally and helps Claude Code sift search hits and open files to find code with less effort.