Paste a repo or app link. It runs on a throwaway cloud machine holding decoy wallets, and you get a verdict and a public evidence page before you open it yourself.
Run it here firstThrowaway cloud machineDecoy wallets insidePublic evidence pageFirst reporter is paidWorks without the tokenRun it here firstThrowaway cloud machineDecoy wallets insidePublic evidence pageFirst reporter is paidWorks without the token
Fig. 02 · The laptop that holds your keys
[01] The take-home trap
A recruiter you have never met sends a coding test. The repo looks normal. An install script can do more than the brief says. It runs on the laptop that holds your keys.
Before you run npm install, paste the repo into Dyepack. It runs on a throwaway machine next to decoy wallets. If the decoys get swept, the evidence page goes public.
Fake job-interview code is one of the top security conversations in crypto right now. Engineers are sent code by strangers all the time, and almost nobody checks it first.
Security tooling had a loud 2026. In May, Vitalik Buterin's post on AI-assisted formal verification landed the day before a $77M exploit at Echo Protocol. Formal proofs cover contracts. Dyepack covers a smaller gap: the code a stranger sends to one developer.
Launch windowNow to Q1 2027
[03] The room
A sealed test room
Every sample gets a fresh cloud machine that exists for one run. It installs, builds and starts the code as the instructions say. Then the machine is thrown away.
Sample[01]
The code a stranger sent
A take-home test, a demo build, an app link. Paste the link instead of cloning it onto your own machine.
Machine[02]
One fresh machine, one run
It exists for one run. Install, build and start happen there, as the instructions say. Then it is thrown away.
Decoys[03]
Decoy wallets inside
The machine holds decoy wallets funded with a few real dollars on the major chains developers use. Honest code has no reason to touch them. If they are swept, the theft is recorded onchain and can't be denied.
Evidence[04]
A public evidence page
Every sample checked gets its own page: the sample hash, what the run did and whether the decoys moved.
[04] A run, drawn as a plan
Sheet 02 · Plan · Not to scale
Paste the code a stranger sent you.
It runs on our machine, not your laptop.
If the decoys move, the page goes public.
01Install scripts
02Decoy wallets
03Network out
Run logCHECKING...
Running on a fresh machine. Decoy wallets loaded. Addresses watched.
[SCROLL TO CONTINUE]
[05] How it works
Paste, run, watch, publish
Paste a link, and a fresh machine installs and starts the code next to the decoys. The run is logged and the decoy addresses are watched. The verdict goes to you and the evidence page goes public.
[01]PastePaste a public repo or app link.
[02]RunA fresh machine installs and starts the code next to the decoys.
[03]WatchThe run is logged. The decoy addresses are watched.
[04]PublishThe verdict goes to you. The evidence page goes public.
Example run · placeholder values
F-01Decoy address moved
F-02Outbound call on install
Dyepack · Verdict<sample>
Finding #01
Decoy wallets were swept
During install, a script read files on the machine. Two decoy addresses moved onchain after that.
Next step
Do not run this on your own machine.
Link the evidence page when you report it.
Evidence pagePublic
Sample
0x…<sample hash>
Source
<public repo link>
Machine
Fresh · thrown away after the run
Run
Install · build · start · network log
Decoys
0x…<address> · watched
Verdict
<what the run observed>
Reporter
0x…<wallet>
[06] Public evidence page
One page per sample
Every sample checked gets its own page. Exchanges and security teams can link to it and act on it. It reports what the code did and nothing more. It names no person or company.
The sample hash, committed at upload.
What the run did: installs, files opened, network calls.
Whether the decoys moved, with the onchain record.
A stable link that exchanges and security teams can cite.
[07] Token · title block
$DYE funds the bait
The tool works without the token. The token pays for bait and reporters. It does not gate the tool.
Ticker$DYE
TypeUtility · community reward
ChainBase
LaunchFair launch, bonding curve
PresaleNone
TeamSmall vested share
InflowsCreator fees · USDC feed subscriptions from security teams
Holding itFunds bait and reporters. The rest is public float.
[08] First reporter is paid
Credit is enforced onchain
01
Each upload commits the sample hash and the reporter's wallet onchain, on Base.
02
The bait addresses issued for that run are registered against it. Decoys sit across the major chains developers use.
03
If any of them is swept later, a keeper watching those addresses releases the reporter's reward automatically.
[09] Who it's for
Who it's for
Developers who are job-hunting or get code from strangers in Telegram groups and hackathon Discords. Security teams at exchanges and protocols who want the evidence feed.
Fig. 03
You got a take-home test from a recruiter you can't verify.
Someone in a dev group sent a demo build to try.
Your team wants a feed of samples it can link to and act on.
[10] Risks, stated plainlyDoes a clean result mean it is safe?
No. Careful attackers may recognise the sandbox and stay quiet. A clean result means nothing was observed on that run, and the verdict says exactly that. It is not a guarantee.
What does a verdict prove?
It reports what the code did on one fresh machine: what it installed, what it opened, where it connected and whether the decoys moved. Verdicts have to be explicit about what a clean result does and does not prove. That wording is still being worked out.
Who pays for each run?
Running untrusted code safely costs real money per check. Creator fees and security-team subscriptions are meant to fund bait and rewards. How the per-check cost holds up at scale is not settled yet.
Do I need $DYE to use it?
No. The tool works without the token. The token's job is to fund bait and pay a crowd of reporters.
What if the repo is private?
The run needs a link it can clone. If it can't, you will see: "Couldn't clone this repo. It is private or the link is dead. Paste a public link instead."
Does the page name the sender?
No. The evidence page reports what the code did and nothing more. It does not name a person or company as the attacker.
[01] ENTRY
[02] DECOYS
[03] NETWORK
Fig. 02 · The laptop that holds your keys
