How to use Claude to run parallel PR workflows
Claude Code coordinates implementation lanes, review and CI before a lander merges and deploys; the author handles decisions and periodically checks the product.
news.ycombinator.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Coordinate [ready task queue] with beads and Claude Code. Use isolated Git worktrees for implementation, agent review and green CI before verification. Rebase verified work before merging and deployment; return complex conflicts or failing checks to the lane and surface human decisions.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- 404softwarelabs
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The author reports 50–100 PRs per day depending on their ability to resolve questions, and identifies CI and human attention as bottlenecks.
- Human involvement
- The author does not review code or individual tasks; they resolve attention requests, periodically review product behavior and submit bugs.
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
- Ran ready tasks through implementation worktrees and review agents.
- Created pull requests and marked lanes verified after green CI.
- Rebased verified tasks, checked master, merged and deployed.
The author describes one Claude Code session coordinating several development lanes, with beads tracking tasks. Each lane takes a ready task, runs a designer when UI changes are involved, and sends implementation work to an agent in a Git worktree. A review agent examines the result. Rejection sends it back through implementation once; approval leads to a pull request.
A lane becomes verified only after CI checks, including tests and vulnerability audits, are green. A lander in the same session then rebases verified work onto master, checks that master is green, merges and deploys. Complex conflicts or failing CI after rebasing send the task back into the pipeline.
The author does not inspect code or verify individual tasks. Instead, another session surfaces decisions or manual actions requiring attention, while dedicated sessions support larger planning and research. They periodically check the product and report bugs through their own session-replay tool. Reported delivery is 50–100 pull requests daily, dependent on resolving questions; CI required a dedicated server. Who it's for: developers exploring concurrent agent delivery with automated checks and a separate queue for human decisions.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · 404softwarelabs
Frequently asked questions
What happens when rebasing causes problems?
The lander returns tasks to the pipeline when conflict resolution is complex or CI fails after rebasing.
What does the human still review?
The author handles items requiring decisions or manual actions, periodically checks product behavior and submits bugs, but does not review individual tasks or code.
More details
Remaining steps
- Reviewed attention requests and periodically checked product behavior.
About this account
Claude Code coordinates multiple development lanes and a lander that rebases, merges and deploys verified tasks.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=49462511. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: 404softwarelabs on Hacker News
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude Code · beads · Git worktrees · CI · session-replay.com
Agent workflows
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.