CKB script review
Attack your script before someone else does.
Give us the code hash of a lock or type script on CKB. We take a real transaction that uses it and try thousands of changes an attacker could make: moving value, swapping cells, editing witnesses, faking proofs. You see which ones your script let through.
01 · Free
Automatic scan
For any script already used on mainnet or testnet. The result is shown only to the key that holds the script's code.
Start a scan02 · By our team
Manual review
For scripts not deployed yet, types with their own rules, multisig owners, or a full report. We read the code and write the attacks for it.
Ask for a review03 · Always on
It keeps learning
Every case teaches the engine. AI agents study what it could not judge and propose new attacks; each one is proven on test scripts before it joins.
See what it learnedAutomatic scan
Paste your script's code hash, an address that uses it, or its page on the explorer. We find a real transaction that uses it and try every attack the engine knows on it.
Only the key that published the script can open the report. Anyone else sees that the scan ran, never what it found.
Scans run in order while our engine is on, usually the same day.
- In the queue
- Running
- Done
Open the report
Your key never leaves your machine: you sign a one-time message, we check the signature.
1 · Run this with the key that published the script
Using CCC instead of ckb-cli
Put the private key in KEY and run, in a folder with @ckb-ccc/core installed:
2 · Paste the signature it prints
This script's code is held by a key nobody can sign with, a multisig or another script, so the report cannot be opened here. Ask for a manual review and we will check with you that it is yours.
Manual review
We read your code, write the honest transactions and the rules your script must keep, and run the engine with them. You get a report and, for each problem, a test that shows it and passes once it is fixed.
We answer by email with what we would look at, how long it takes and the price.
How it learns
Every scan is kept as a case. When the engine cannot replay a transaction, or finds something it cannot judge, we study the case with AI agents, write the missing attack or setting, and prove it on test scripts built with and without the flaw before it joins the engine.
The list is the engine's own history, newest first.