I’m giving away 10 GTD WL spots on this post
> Follow @le699a, @srhsnn & @elysium_machine
> Like & RT their latest post
I’ll pick the winners in a few hours
Kite's catalog said an endpoint costs $0.001.
My agent paid $0.002 through Kite Passport, that's double the listed price. Nothing warned me.
Passport caps how much an agent can spend. Nothing checks what it is actually paying for.
So I built Signoff for SERV Hackathon 02 (Kite track) , it checks the payment before Passport signs it.
What I found on Kite's own catalog:
I sent unpaid requests to a 343-endpoint sample of the catalog on Nov 3, and again on Nov 4, and compared each live x402 payment request with its listing:
→ 24 of 165 endpoints marked "works" asked more than their listed price. Mostly 2×, up to 10× ($0.10 against a listed $0.01).
→ Endpoints marked "untested" asked up to $10 against a listed $0.022.
→ One "works" endpoint answered 410 Gone.
→ The catalog records no expected payee, so Passport pays whichever address the live 402 names.
Small amounts today. The point is structural: nothing between the listing and the signature compares them.
How Signoff works
1. Send the request unpaid and read the live terms: price, payee, asset.
2. Compare them with the listing. Refuse on price drift, an over-budget price, a payee that changed since first use (pinned like SSH known_hosts), a retired endpoint or the wrong asset.
3. Ask SERV Reasoning whether this endpoint, called with this exact request, serves what the user asked for.
4. Only on ALLOW, pay through the user's passkey-approved Passport session.
5. Ask SERV whether the reply actually answered, and write a receipt for every decision.
The SERV rule I like most: every yes or no must quote verbatim evidence from the request or reply. If it can't, the verdict becomes "unsure" and the payment is held. It fails closed instead of guessing.
Live: asked for "weather in Lagos", the agent's request pointed at Berlin. SERV refused, quoting "lat=52.52".
Results
→ The same call Passport paid at 2× was refused before any money moved.
→ 3 paid ALLOW calls on Base mainnet, each with a receipt.
→ Receipts for 5 refusal types: price drift, over budget, retired, off-intent, payee change (the last shown with a hand-edited pin; no real payee change was observed).
Try it free, no wallet:
npm run guard -- report
There's also a drop-in agent skill, so any Passport agent pays through the guard instead of calling kpass directly.
I've sent the findings to Kite.
Next: asking Kite to let Passport accept the exact price and payee Signoff approved, so nothing can change between the check and the signature.
@openservai@GoKiteAI
➥ @elysium_machine NFT Giveaway
▰▰▰
We're giving away:
- 30GTD
> 4,444 ERC-404 machines on @Enter_Elysium
> every machine starts dark, consecrate it and it starts printing tokenized stocks
To enter :
- Follow @elysium_machine & @iamlequanghuong
- Like + retweet Elysium Machines pinned post and this one
- Gib EVM walle
Winners will be selected in 36 hours.
Good luck everyone
Elysium Machines — launch date is set.
Friday, Nov 6 · 15:00 UTC
4,444 machines, one token: $MACHINE (ERC-404).
— Hold 1 $MACHINE → a machine appears in your wallet
— Drop below 1 → it dissolves back into the pool
— Consecrate it → the token burns, the machine lights up and farms HYPE & tokenized stocks
— Lit machines share in the fees
3% fee on every $MACHINE trade.
5% fee on NFT.
Fair launch:
— No presale, no team allocation
— 100% of supply in one locked pool
— Everyone buys on the same curve
WL wallets gets priority access to the system.
WL closes in 24h: https://t.co/tE4vIIpOMx
Docs: https://t.co/1A0aJHdoFV
Systems check complete. See you at boot.
Elysium Machines just generously gave me and my community 15 WL spots of their upcoming NFT (I will keep 2 for me)
@elysium_machine is 4,444 ERC-404 machines on @Enter_Elysium which mint on friday, Nov 6 · 15:00 UTC
Rules is simple:
- Like + Retweet this tweet and their latest tweet
- Follow me and @elysium_machine
I will hand pick the winner in 12 hours
@ilblackdragon Mr Illia , i asked a builder inclined question on your previous post will be nice if you had a look, i also dmed you the question some days back
@ilblackdragon@NEARProtocol@near_intents@near_ai would you publish a model-vs-model comparison on your own eval set if the set were committed on-chain before the run, and anyone could verify the scores without seeing the questions?
Trying to find out if that constraint is real before I build it. What would make you say no?
@ilblackdragon@NEARProtocol@near_intents@near_ai I'm building Fūin for NEAR HACKATHON (privacy track): publish a model benchmark score that anyone can verify without seeing the test set. The set is
committed on NEAR before the run, the models run in NEAR AI Cloud TEEs, and the certificate checks offline with no account
secured @slymzcrew GTD FREEMINT spots
1,527 slime nfts coming on Elysium
How to enter:
Follow me & @slymzcrew
Like and repost this post
Comment your EVM wallet address
Winners will be picked randomly and announced in 24 hours.