Engineering role
AI Code Reviewer
An AI code reviewer that reads the whole pull request and tells you what it breaks, not just how it looks.
Give it a pull request and it reads the diff, the files around it and the history behind them, then writes one complete review: what the change does, what it might break, and which comments are blockers versus suggestions. It is built for the judgement a linter cannot make — a missing auth check, an N+1 query, an error path nobody tested.
Things you could ask for
Written the way you would actually say them. The agent plans the steps itself.
- “Review pull request #214. Lead with anything that is a blocker, then suggestions, then nits — and say what the description leaves out.”
- “Read this diff and tell me what could break in production that the tests would not catch.”
- “Check every endpoint this PR touches for missing input validation or authorisation checks, and quote the lines.”
- “This PR changes the billing code. Walk through what happens to a customer mid-cycle, step by step, before and after.”
- “Every morning, review whatever was opened yesterday in the main repository and leave me one summary per pull request.”
What you get back
One review per pull request: a short summary of what the change does, then findings grouped as blockers, suggestions and nits, each tied to a file and line with the reason it matters and a proposed fix. Delivered in chat, or on a schedule as a notification you can read on your phone.
Tools it works with
A role is a job description; connectors are what let it do the job on your real work instead of on whatever you paste into a chat.
Where it is the wrong tool
- It does not merge or push. The GitHub connector covers reading code, issues and pull requests; your branch protections and a human approval stay the thing that decides what lands.
- A remote agent cannot run your code — it has no shell, so it reasons about behaviour from the source and will miss what only shows up at runtime. Keep your test suite and CI as the ground truth.
- It is not a linter. Formatting, import order and style rules are cheaper and more consistent in a tool built for them; point the reviewer at logic.
- On a very large diff, split the review by area. A 4,000-line pull request gets a shallower read than four 1,000-line ones.
Starting an agent with this role
- 1Create a Burrak account — any plan works, and your first week is $1.
- 2In the Marketplace, open the Roles tab, find Code Reviewer and choose “Start an agent as Code Reviewer”. That creates a remote agent with the role’s instructions in its system prompt.
- 3Connect the tools it needs — GitHub, GitLab, Sentry, Linear — from Connectors.
- 4Give it a brief. It keeps working in the cloud after you close the tab, and you are only charged while it is actually working.
The Code Reviewer role, in practice
- How is this different from asking a chatbot to review code?
- Context and follow-through. Connected to GitHub, the agent reads the pull request itself and the surrounding files rather than a snippet you pasted, and it can go and look at the commit that introduced a function before it comments on it. It also runs unattended — on a schedule, over whatever was opened since the last run.
- Will it post comments on the pull request?
- It writes the review back to you — in chat, or as a notification when it runs on a schedule. Read it, keep what is right, and post it yourself. That is deliberate: an automated reviewer that comments directly trains a team to ignore it the first time it is confidently wrong.
- Does it see my private repositories?
- Only what you authorised on GitHub’s consent screen, and only when a brief asks it to look. The grant lives on GitHub’s side, so you can review or revoke it from your GitHub settings at any time.
Capability reference
Summarised from the role’s instructions, which are adapted from the agency-agents collection (AgentLand Contributors), used under the MIT licence.
- Reviews for correctness, security, maintainability, performance and test coverage — not tabs versus spaces
- Marks every finding as a blocker, a suggestion or a nit
- Is specific: names the line and explains why, rather than “security issue”
- Suggests rather than demands, and calls out good code as well as bad
- Gives one complete review instead of drip-feeding comments across rounds
Your AI assistant is waiting.
Put it to work today.
Put an agent on the work that never needed a human in the first place — the research, the reports, the follow-ups — and take the week back.
$1 for your first week · Cancel anytime · Works on Mac, Windows, iPhone, Android & Web