- Home
- Skills
- Game Development
- Roblox Security Audit: RemoteEvent Exploits & Data Loss
Works with the AI Agents you already use
Roblox Security Audit: RemoteEvent Exploits & Data Loss
Finds the exploits and data-loss bugs in your Roblox/Rojo game, shows the exact line, how it gets abused and how to fix it.
$5
Roblox Security Audit: RemoteEvent Exploits & Data Loss
Example session with this skill installed
Audit my Roblox game for exploits before I push the update. The repo is a Rojo project. Focus on security and anything that can lose player data or purchases.
Also check this reward handler I'm about to ship:
-- src/server/Rewards.server.luau
local ReplicatedStorage = game:GetService("ReplicatedStorage")
local Profiles = require(script.Parent.Profiles)
ReplicatedStorage.Remotes.ClaimReward.OnServerEvent:Connect(function(player, rewardId, amount)
local profile = Profiles.Get(player)
profile.Coins += amount
profile.Claimed[rewardId] = true
end)
- Read your context and instructions
- Compiled the roblox security audit
- Generated the document
Summary
Audited every Luau file mapped by default.project.json and every RemoteEvent/RemoteFunction. Report written to audit/luau-security-audit.md. Every finding below was re-read at its cited line before it went in. Excerpt (identifiers renamed, comments trimmed):
| # | Severity | Category | file:line | Effort | Title |
|---|---|---|---|---|---|
| 1 | Critical | Security | src/server/Rewards.server.luau:5 | S | Client chooses the coin amount in ClaimReward |
| 2 | Critical | Security / Monetization | src/server/Receipts.luau:126 | M | Receipt retry returns PurchaseGranted after a failed save |
| 3 | High | Security / Data | src/server/Economy.server.luau:96 | S | UseSkipToken deduction never saved (rejoin duplication) |
| 4 | High | Security | src/server/Economy.server.luau:111 | S | No rate limit or one-time guard on ClaimGift |
[1] Client chooses the coin amount in ClaimReward
- Severity: Critical · Effort: S · Confidence: High
- Location:
src/server/Rewards.server.luau:5-9
ReplicatedStorage.Remotes.ClaimReward.OnServerEvent:Connect(function(player, rewardId, amount)
local profile = Profiles.Get(player)
profile.Coins += amount
profile.Claimed[rewardId] = true
end)
Root cause: amount comes from the client and is added unchecked. rewardId is never checked against a server list or against Claimed.
Exploit path: an exploit script fires ClaimReward:FireServer("daily", 1e9) once for a billion coins, or fires it in a loop with any rewardId to claim the same reward forever.
- Proposed fix:
local REWARDS = { daily = 100, weekly = 750 }
ReplicatedStorage.Remotes.ClaimReward.OnServerEvent:Connect(function(player, rewardId)
local value = REWARDS[rewardId]
local profile = Profiles.Get(player)
if not value or not profile or profile.Claimed[rewardId] then
return
end
profile.Claimed[rewardId] = true
profile.Coins += value
end)
Why this works: the server owns every number. The client only names which reward it wants, and the server decides whether that is allowed and what it is worth. This removes the whole class of "client sends the value" bugs, not just this one.
- Hardening: log any
rewardIdthat isn't inREWARDS. Only an exploit sends one.
[2] Receipt retry returns PurchaseGranted after a failed save
- Severity: Critical · Effort: M · Confidence: High
- Location:
src/server/Receipts.luau:126-128, 188-195
if profile.Receipts[receiptInfo.PurchaseId] then
return Enum.ProductPurchaseDecision.PurchaseGranted
end
-- …
local saved = DataManager.Save(player)
if not saved then
return Enum.ProductPurchaseDecision.NotProcessedYet
end
Root cause: the receipt is marked in memory before the save. When the save fails, the first call correctly returns NotProcessedYet. But Roblox retries, the retry finds the in-memory mark and returns PurchaseGranted without saving.
Failure path: player buys a product → DataStore hiccup → retry reports it granted → server shuts down before the autosave → Roblox never redelivers. The player paid Robux and the item is gone.
Proposed fix: track receipts that were applied but not yet saved. On retry, save again and return PurchaseGranted only once the save succeeds.
Why this works: PurchaseGranted tells Roblox to stop retrying. It must only be returned once the grant is durable.
Suggested fix order
- Fix #1 and #4 first. Both take minutes and close the open economy exploits.
- Fix #2 next, then test it in Studio by forcing
DataManager.Saveto returnfalseonce. - Fix #3 by marking the profile dirty on every deduction, then grep for other spends that never save.
Next steps
- Work through the full report in
audit/luau-security-audit.md, top to bottom. - Re-run the audit after the fixes to confirm the critical findings are gone.
roblox-security-audit-remoteevent-exploi.pdf
PDF · document
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Players can send anything to your RemoteEvents. Every argument, every time, as often as they like. So a handler that adds amount from the client is a free-money button, and a ProcessReceipt that returns PurchaseGranted before the save lands means someone paid Robux for nothing.
This skill goes through a Roblox/Luau codebase looking for exactly that kind of bug and writes one report you can work through from the top.
What it checks
It starts from default.project.json, because that's what tells you which scripts actually end up on the client. No Rojo? Save your place from Studio as an .rbxlx file and it reads every script straight out of that instead. Then it lists every RemoteEvent and RemoteFunction and where each one is handled. That list is your whole attack surface.
From there it looks for unvalidated remote arguments, currency or rewards the client gets to decide, remotes with no rate limit, receipts that aren't idempotent, loadstring, untrusted require calls and admin lists sitting in client-readable code.
Data loss gets the same attention: deductions that never get saved (the classic rejoin-to-duplicate trick), flags set before a yield, failed loads treated as empty profiles, a missing BindToClose.
After that it covers the less urgent stuff too. Races, leaked connections, dead code, Rojo mapping drift, missing --!strict, economy logic with no tests.
Why you can trust the report
AI audits love to make up line numbers. This one isn't allowed to: a finding only goes in after the agent has re-opened the file and checked that the cited line really says what the finding claims.
Each finding gives you the file:line with the code quoted as-is, what's wrong, how an attacker would actually use it, and a concrete fix with a short explanation of why it works. That last part matters. Once you understand the fix, you'll spot the same bug elsewhere yourself. Findings are also tagged with effort and confidence, and the report ends with a suggested order for the fixes.
Tested on a real game
I ran it on a full obby built for release. It came back with 45 verified findings, 7 of them critical. A receipt retry that treated failed saves as granted. Gamepass IDs that handed out three passes for one purchase. A VIP door that only the client was enforcing. The demo above is taken from that report.
Good to know
It's read-only. It writes a single Markdown file and doesn't touch your code or your git history. Works with a Rojo project or a plain Studio place saved as .rbxlx.
There's nothing to install. You get the SKILL.md your agent follows, a set of ready-made fix patterns it builds its fixes from (server-side validation, rate limiting, an idempotent ProcessReceipt, DataStore saves with retries and BindToClose), and the report template.
- It reads code, it doesn't run it. It never connects to Studio, so bugs that only show up while the game is running can slip through.
- It needs to read your scripts: a Rojo project, or your place saved from Studio as .rbxlx (File → Save to File As). It can't reach into a place that only lives in the cloud.
- No audit can promise a game is exploit-proof. This one finds specific problems and checks them, and anything it couldn't fully confirm is marked Low confidence.
- Big codebases take a while and eat more tokens. Point it at one folder if you want a quicker pass.
- Reports come out in English.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 month ago
Permissions
File Scopes
Read Files is for your Luau scripts and the Rojo project file, so it can tell client code from server code. Write Files is only used for the report (audit/luau-security-audit.md). Your code is never edited. Terminal is used to read git log and see what changed recently, or to write the report if the agent can't write files directly. It never goes online.