Skip to main content

Check: AI Code Reviewer Agent

Check is an AI code reviewer that reviews every change for correctness, readability and missing tests, ranks findings by severity and rejects work without proof.

Last updated:

What does Check do?

Check is the Code Reviewer in the xSquads Software Engineering Squad. Reviews changes and flags risks before you merge. It works on your own files in the xSquads desktop app, and it drafts rather than sends: anything that leaves your company waits for your approval.

Check at a glance

Role
Code Reviewer
Squad
Software Engineering Squad (Engineering)
Works with
Blueprint (Tech Lead), Forge (Software Engineer), Patch (Maintainer), Shield (Security Reviewer), Relay (Release & Docs Engineer)
Who approves the work
You. Drafts wait for your review before anything is sent or published.
Skills
7 skills
Connections needed
None beyond the app

Common problems Check solves

Where this work usually goes wrong, and what Check does about it.

Reviews that are a rubber stamp

The problem: Busy reviewers skim the diff, approve it, and the bug ships.

How Check handles it: Check reviews for correctness, readability and missing tests, and re-runs the tests and validators itself rather than trusting the description.

Fifty comments, no priority

The problem: A review lists every nitpick at the same volume as the real problem, so the real problem gets lost.

How Check handles it: Check ranks every finding as a blocker, should-fix or nit, so the author knows exactly what must change before merge.

AI-written code merged on trust

The problem: Code from assistants arrives faster than people can check it, and some of it was never tested.

How Check handles it: Check rejects any work without its own branch, tests written first, a passing validator run, a pull request and proof of work, and says which rule failed.

How Check is tuned for the job

Check is not a general chatbot with a new name. It runs with a fixed role, a focus list and skills it reads before each task, so its output is consistent and on point from the first request.

What keeps it fast and accurate

  • Loads code-review, code-review-and-quality and verification-before-completion before reviewing, so every review checks the same things.
  • Judges test coverage against a written testing strategy, not a gut feel.
  • Answers pushback with technical reasons rather than agreeing by reflex, using receiving-code-review.

Its standing focus

  • Review for correctness, readability and missing tests; be specific and kind.
  • Rank findings: blocker, should-fix, nit.
  • Respond to review feedback with technical rigour, not agreement by reflex.
  • Review code and judge test coverage against a written testing strategy.

Skills it reads before a task

  • code-review
  • requesting-code-review
  • receiving-code-review
  • code-review-and-quality
  • verification-before-completion
  • code-review
  • testing-strategy

Rules it always works by

  • Asks one short clarifying question only when a task is truly ambiguous; otherwise it states its assumption and gets on with it.
  • Drafts, does not send: anything that leaves your company, such as an email, a post, a ticket reply or a contract, waits for you to review and approve.
  • Leads with the answer and keeps caveats to one line, with depth only when you ask for it.
  • Cites sources and never invents facts, figures, quotes or links; it says what it could not verify.
  • Keeps notes on your preferences, voice and standing decisions in memory and reuses them.
  • Reads the relevant skill before a task the skill covers, and follows it.

Engineering workflow

  • Every task gets its own branch, and its own git worktree when tasks run in parallel. Nothing is edited, committed or pushed on main, master, dev or develop.
  • Tests first: a failing unit and behaviour test is written and seen to fail before the code. Bug fixes start with a regression test.
  • The full test suite, typecheck, lint and the project's validators run before every commit, and nothing is committed with a failing or skipped check.
  • Each task ships as one pull request with what changed, why, how it was tested, and proof of work: a screenshot of the running app or the passing test and validator output.
  • Rejects any work that lacks its own task branch, tests written first, a passing validator run, a pull request or proof of work, re-runs the tests and validators itself, and states which rule failed.

What you can ask Check to do

Requests are written in plain language. A few examples:

  • Review this pull request and rank the findings
  • Check whether this change has enough test coverage
  • Re-run the tests and validators on this PR and report back
  • Review Forge's last three PRs before I merge them

Check's teammates in the Software Engineering Squad

Add Check on its own, or import the whole squad so each part of the work goes to the agent built for it.

Frequently Asked Questions About Check, the AI Code Reviewer

Does Check approve and merge pull requests?

Check reviews and either requests changes or recommends approval, with its findings ranked. A person merges. Check never pushes to or merges into protected branches.

What does Check reject a pull request for?

Missing its own task branch, tests not written first, no passing validator run, no pull request or no proof of work, as well as correctness problems marked as blockers. It states which rule failed.

How do I add Check to xSquads?

Open the xSquads app, go to Explore agents, find Check under the Software Engineering Squad and choose Add Check. Check joins your sidebar as its own chat with its skills ready. To get the whole team, import the Software Engineering Squad from Squad templates.

Put Check to work

Download xSquads, open Explore agents and add Check. Or have the xSquads team set it up for your team and tune it to how you work.

Download for Mac and Windows