Urgent.News

What's breaking now, across hundreds of outlets.

AI

Live Coding with an AI Team: A Safe Way to Review a PR (One Writer, One Reviewer, One Publisher)

I ran a small pull request review live with AI agents, and the part that went wrong was never the code. It was the handoff : two agents editing one folder, a bot "reply" that proved nothing, a caption that said "checks passed" while CI was still queued. So I turned the setup into a rule you can follow on stream or at your desk: One issue. One independent check. A human decides what ships. Below…

In a live coding session, the author employed an AI team to review a pull request (PR). The main issue that arose was not with the code itself, but rather with the handoff process between the AI agents. Two agents were editing the same folder simultaneously, and a bot's "reply" did not provide any meaningful information. A caption stating "checks passed" was displayed, even though the continuous integration (CI) queue was still active.

To avoid such complications, the author established a set of guidelines for reviewing a PR using AI agents.

The author recommends the following steps:

1. Assign a specific task to each tool. The tool must not perform multiple functions. For instance, Claude is responsible for writing tests and the small fix, while OpenCode reviews the same diff without making edits.

2. Ensure that each author has their own Git worktree, which is a separate checkout of the same repository. This prevents interference between different agents working on the same folder.

3. Review the exact changes that will be published. The reviewer should have a saved diff to work with, preventing them from writing comments about code they did not create. The reviewer should ask for evidence rather than opinions when examining the diff.

4. Verify connections one at a time. Slack mentions, Orca bot handoffs, and Hermes messages are separate surfaces, and granting access to one does not automatically provide access to the others. The author suggests testing one human mention and one reply before concluding.

5. Designate a single publisher for PR comments. Duplicate PR comments can result from repeated triggers or multiple bots posting the same result. Keep early reviews in local notes or a single Slack thread, and give publishing rights to a single identity tied to a specific commit.

6. Clearly state the CI state. The CI check should describe the code tested in that check. After the merge, examine the workflow for the merge commit on the main branch to determine if the run exists, has started, has executed, and if any required jobs have passed.

7. Stream live with a single window focus. Ensure that the audio is clear, and avoid showing sensitive information such as tokens or private repo names.

The author has compiled these guidelines into a 41-page visual PDF, which is available for free at the provided link. The author also runs live demonstrations of AI agent workflows on YouTube, which can be found at the provided YouTube link.

Rewritten by urgent.news; may contain errors — check the original.

Source: dev.to

Read the original at dev.to →

Report a problem with this story

More in AI

AI衝擊教育 黃昆輝:培養思考判斷創新的善良國民

記者林曉雲/專訪 生成式AI時代,學校要把孩子培養成什麼樣的人?黃昆輝教授教育基金會董事長黃昆輝表示,科技快速改變教與學,但教育本質仍是要培養能思考判斷創新的善良國民。 黃昆輝提到教育哲學學者保羅・奈許,並以物理學家與水管工反問,一位知識淵博卻自私冷漠的物理學家,和一位待人溫暖的水管工,…

More from Sunday 22 November →