█

AI/TLDR

Anthropic · 2026-11-16 · notable

How the Claude Code team verifies agent code — tests, reviewers and gates

In short

Addy Osmani explains how the Claude Code team checks code written by Claude: end-to-end, property and differential tests, a tuned reviewer, and a commit gate that blocks unverified changes.

Anthropic · 2026-11-16 · 2 sources checked

Claude blog card for the post on how the Claude Code team thinks about verification

Anthropic's Claude Code team shares the checks that let it run many agents at once without reading every line.

What is it?

The Claude Code team's verification playbook, written by Addy Osmani on the claude.dev blog, explains how the team trusts code that Claude writes and reviews. The team says about 90% of its code is written and reviewed by Claude, so verification, not manual review, is what lets it run dozens of agents in parallel.

How does it work?

End-to-end tests come first, then property tests (fast-check, Hypothesis, proptest) and differential tests that run the old and new system on the same inputs. A `verify` skill in `.claude/skills/` makes Claude run the app like a QA engineer, a 39-item reliability checklist runs only the 3 to 10 items that match a change, and one reviewer is tuned with a `REVIEW.md` read from the base branch so a PR cannot change its own checks.

Why does it matter?

Teams moving to agent-written code get a concrete recipe instead of a slogan: which test types are hard for an agent to fool, how to roll out a commit gate in observe mode before enforcing it, and how to measure reviewer recall. The post reports that a separate rubric grader raised task success by up to 10 points in internal tests on structured file generation.

Who is it for?

engineering teams using coding agents

Try it

Add a verify skill under .claude/skills/ and a REVIEW.md for Claude Code's reviewer

Sources · 2 outlets

Tags

  • claude-code
  • anthropic
  • verification
  • testing
  • code-review
  • coding-agents
  • property-testing
  • differential-testing
  • best-practices

← All releases · Learn AI