Built for Build Before Q-Day. Live on the QoreChain testnet, qorechain-diana. Read the source
Security

What Fenestella protects, and what it does not

The threat is simple: someone has your account's classical (ECDSA) key and nothing else. A quantum computer could recover it from the public key; malware or a leaked file can do it today.

Enforced by the chain

What stops that attacker

The chain, not this site. On QoreChain:

  • Every native transaction needs the ML-DSA-87 signature as well. The classical key alone is refused.
  • Only the post-quantum key can open, widen or renew an EVM window.
  • An EVM transaction outside an open window, or outside its scope, is refused.

We ran this ourselves on testnet. With only the classical key of a protected account, a native transfer, an open-window message and an EVM transfer were each refused, and the balance did not change. You can run those three, and 10 more against the agent, the vault, the seal and the AI, in the attack lab.

Three layers

What stops a stolen agent key

  1. The chain. The agent's key is linked to your account with a spending rule the chain enforces: it may call one function, AgentVault.payBill, and move at most 0.000001 QOR of your own balance.
  2. The vault. It pays only payees you approved, within limits you set, from the budget you put in. Only your ML-DSA-87 signature, checked on chain by pqcVerify, changes payees or limits, pauses the vault or withdraws.
  3. The AI. It reviews every bill, but it never has the last word. A guard written in code can only make its answer stricter, and the chain and the vault come after both.
Know this before you approve

What an open window still allows

  • Anyone holding the classical key can use it, within its bounds, until it runs out or you close it.
  • The cap counts native QOR only. Token amounts are not capped.
  • A token approval made through a window stays in force after the window closes.
  • A window with no targets reaches every contract.

Fenestella shows these facts for your account and steers you to the narrowest window that does the job.

What Fenestella itself can and cannot do

It can
  • Read public chain data about any address
  • Hand your wallet a message to sign
  • Show a dApp's request with the origin your browser reports
  • Run a testnet agent and a relayer, with their own throwaway keys
It cannot
  • See or store a private key or recovery phrase when you use QoreX
  • Sign, broadcast or change anything without your approval in the wallet
  • Let the dApp choose the wording of what you approve
  • Move your money outside the leash you signed: one vault function, approved payees, your limits
Honest limits

Limits you should know

  • This is a hackathon project. It has 54 contract tests and 91 unit tests, and its contracts went through two QoreChain AI audits with every finding answered in AUDIT.md. No human firm has audited it.
  • The AI can be fooled, and a payee you approved can still bill you for real amounts within your limits. Approve only payees you trust.
  • A wrong answer from this site could mislead you. The exposure logic is open, unit-tested and treats anything it could not read as unknown, never as safe. You can check any account against the chain yourself.
  • If a phishing site imitates Fenestella, your wallet's own approval screen is the last check. Read it.
  • The testnet key option keeps a phrase in the tab's memory. It is for development and refuses mainnet.
  • The contracts have no admin and no upgrade path. Only AgentVault holds funds, each owner's agent budget, and only that owner's ML-DSA-87 signature can take it out. The vault's admin actions and FenestellaReceipts never trust msg.sender: they require a signature checked by the chain's precompile.

Report a problem

Open an issue on GitHub. For anything that puts users at risk, say only that you have a security report and ask for a private channel.

Attack lab · FAQ · Docs