# Claude Code team lets low-risk PRs merge on automated checks alone

> The Claude Code team sets human approval by each change's risk and scores its AI reviewer daily against PRs that caused past incidents.

- Status: reported (One outlet, not independently verified yet.)
- Topic: AI in Practice
- Published: 2026-11-12T09:26:58+00:00
- Updated: 2026-11-12T10:10:29.516705+00:00
- Web page: https://ai-news.ikangai.com/story/claude-code-team-lets-low-risk-prs-merge-on-automated-checks-alone/
- Seen in: Demetrios from the AAIF
- Change: other (Other)
- Applies to: Claude Code

## What changed

Low-risk pull requests from trusted authors can now merge on automated review and security checks alone, without human approval.

The Claude Code team now lets trusted authors merge low-risk pull requests on automated checks alone. No human approval is required once the automated review and security checks come back green. The team's engineering blog post, published October 15, 2026 and written by Addy Osmani, Morgan Westlee Lunt and Qing Wang, describes optimizing for the least human review time that doesn't raise reverts or incidents. The rule is set by the change rather than the author. A classifier reads each change and assigns a risk tier. Low covers docs, copy, internal-only features and additive changes behind a flag. Medium covers ordinary product changes and sensitive areas such as caching, the auto-updater and system prompts. High covers the highest-risk code and a small set of paths that always need two people.

The lighter path has conditions. Only authors on the repo's trusted list get it, mostly the team that owns the repo. A second classifier can also ask whether a particular change needs a person, even when its tier says it doesn't. Some paths always need one, including anything owners have listed as sensitive and any CI workflow file. An agent's approval never counts as a person's. The post also says whoever prompted an agent-written change counts as one of its required reviewers. An approval covers one specific commit, so new commits the system can't vouch for require a fresh approval.

The post explains why automated review carries this much weight. Claude reviews every pull request before a person looks. Findings are checked by other agents, and a separate group votes on each one before it is posted. Findings about style, diagnostics or cleanup are dropped. The team says it A/B tested those filters on its own PRs. Every finding has to be answered, and a pull request can't merge while any bug or security finding is still open. The reviewer's quality is measured against a set of past pull requests that caused incidents. The team collected those along with their fix-forwards and reverts, and had engineers confirm them. The reviewer runs against that set daily, and the post says every change to the reviewer is scored against all of it.

The same post also measures the reviewer's noise. The team sorts each comment by how the author responded: whether they fixed it, called it wrong, called the behavior intentional or decided it didn't matter. It also runs the reviewer on PRs known to be clean and counts how often it reports something that would have led to a harmful change. The team's stated priority is catching real bugs first and cutting review rounds second. The post says that when agents are the authors, noise is more expensive, since an agent fixes each nit with another commit and another round. One CLI pull request went through more than fifty rounds overnight before its author stopped it. When the team ran git blame on what the reviewer kept flagging, about 59% was code that had been there since the first pass.

The post reports that changes landing per day in Anthropic's monorepo went from about 2,500 in May to more than 30,000 in September. Merged pull requests to Claude Code's codebase more than doubled from June to September, while the share that had to be reverted fell by more than 40 percent. Claude Code lives in its own repository, separate from the monorepo. The post presents reverts and incidents as the guardrail for the approach. It also says the team doesn't yet have circuit breakers that tighten approval when those numbers move. Engineers there want them before more code moves to zero-human merges. The post's own Code Review announcement is more cautious. It says approving a pull request is still a human call, which the post describes as a line that keeps moving inside its own repos.

## Why it matters

Risk-tiered approval shows how teams can use AI code review to speed up low-risk merges while keeping human checks on riskier changes.

## What you can do

Tier code review by change risk and benchmark AI reviewers against past incidents before relaxing approvals.

## Original source

- [How the Claude Code team does code review / claude.dev Blog](https://claude.dev/blog/solving-code-review-at-anthropic/?nlb=zU6uBwLdqS)

---

AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it.

Cite as: IKANGAI AI News, "Claude Code team lets low-risk PRs merge on automated checks alone", 2026-11-12, https://ai-news.ikangai.com/story/claude-code-team-lets-low-risk-prs-merge-on-automated-checks-alone/
