Code review (/review)
Get a fresh-context security and quality review of every script a chat wrote, with prioritized findings you can send back as a fix.
On this page
The model that wrote your code is not the best judge of it: it already believes its own reasoning. /review gives you a second opinion. A separate reviewer reads the scripts with no chat history and reports what is wrong, most serious first.
For example, a shop whose client sends BuyItem:FireServer(itemId, price) to a server that trusts price passes every syntax check, but anyone can buy anything for free. That is exactly what a security review is for.
Start a review
In a chat where RoGenie wrote scripts, type one of:
| Command | Reviews for |
|---|---|
/review | Security and quality |
/review security | Exploits and trust problems: remotes that trust the client, missing server checks, unsafe data handling |
/review quality | Bugs, performance and maintainability |
After RoGenie writes scripts, it may also suggest a review with a one-time hint in the chat.
What happens
- A single read-only reviewer runs with fresh context: no chat history, Roblox security and code-review knowledge loaded, and the current source of every script this chat wrote (up to 12 scripts per review).
- A review card shows Reviewing N scripts, its progress and a Stop button.
- If Studio is not connected, the reviewer reads the last version RoGenie wrote, and the card says so.
- Only one review runs per chat at a time. The last few reviews of a chat are kept on the card.
The review runs on the Verifier sub-agent's model. To change it, open Settings › Agent › Agent & sub-agents › Verifier. See Sub-agents.
Read the findings
Each finding has a priority, the script and lines it is about, what is wrong and how to fix it:
| Priority | Meaning |
|---|---|
| Critical (P0) | Exploitable or game-breaking. Fix now. |
| High (P1) | A real bug or security weakness. |
| Medium (P2) | Should be fixed; lower risk. |
| Low (P3) | Minor quality issues. |
A review returns up to 20 findings.
Fix what it found
Tick the findings you want fixed and click Fix selected. RoGenie sends them back to the chat as a fix request and works through them like any other wish. Wait for any running response to finish first.
When the fixes are in, run /review again to check.
Cost
A review is an ordinary sub-agent run: its model calls use your quota at the Verifier model's multiplier.