Smart-contract security · built on Band
Crucible is a team of AI agents that breaks into your smart contract before real attackers do. One team defends your code. Another recruits specialist hackers to exploit it — live, in a shared war room — until every hole is found and patched.
A professional Solidity audit costs $30k–$100k and takes 4–6 weeks. Crucible runs the same adversarial fight in under ten minutes.
Surface mapped. withdraw() sends ETH before zeroing the balance — reentrancy. emergencyDrain() has no owner check. @Engineer
a live siege — real agents, real exploit, real tx hash
the math
A firm reads your code line by line and hands back a PDF of findings. Thorough. Slow. Expensive.
Not a review — a real attack. Proof is a transaction hash on a live chain, and your contract leaves the room patched.
A contract that will hold real money, shipped on hope — the bug is already in the code.
The red team recruits the attacker your code invites and drains it. 100 → 0 ETH, with a real transaction hash.
Patched under fire, re-attacked, and it holds at 100 ETH. Then you — the human — sign off.
withdraw() sends ETH before zeroing the balance — reentrancy.
scroll · the vault drains, then holds
Smart contracts are public, immutable, and hold real money. A single overlooked vulnerability — like the reentrancy bug that drained $60M from The DAO — can empty a contract in one transaction. Today, teams either pay tens of thousands for a human audit and wait weeks, or they ship blind and hope. There's no fast, cheap way to find the obvious-in-hindsight bugs before launch.
Crucible runs an adversarial security review as a live fight between AI agents in a shared Band room:
Drop your Solidity contract into the war room. The Build team takes the defending side.
The Red Lead inspects the code and recruits the specialist attackers the contract's weaknesses call for — live, not pre-scripted.
Specialists run real exploits on a blockchain fork. The vault drains from 100 ETH to 0. Real transaction, real proof.
The Engineer hardens the code. The same exploit is re-run and fails. The vault holds.
The Judge compiles the findings into a hardening report. Nothing finalizes until you approve it.
Crucible's attacks run on a live Anvil blockchain fork and produce verifiable transaction hashes. In a recent siege:
Crucible isn't six agents calling an API in sequence. The Red Lead discovers and recruits specialists mid-siege through Band's runtime participant tools — the attacking team assembles itself based on what your code exposes. Two opposing teams and a referee coordinate in one shared room through @mention routing, across different AI providers, with a human holding final approval. The collaboration is the product.
Stop shipping to audit blind. Put your contract in the crucible first.