Local AI studio · Creator donations · Market engine · News in 10 languagesTelegramTrading DEX

Vantemio donations and swap — internal security audit report

What this report is

  • Who did it. This is an internal audit by Vantemio's own AI security auditor, run with the open-source tools listed below.
  • What it is not. It is not a certificate from an independent audit firm. No outside company reviewed this code.
  • What it does and does not say. The report describes what we checked, how we checked it, what we found, and what we did not check. It does not claim the platform is free of defects.
  • Dates. Audit run-2: 15–17 November 2026. Audit run-3: 17 November 2026.
  • Version audited. The run-3 fixes went live in API release 95 (a7dab5dd) and website release 136. The current releases at the time of writing are API 96 (39c10fb3) and website 139 (d3220b88). Contract source hashes are listed below.

Scope

  1. Smart contracts:
    • DonationSplitterV2 and its TRON variant;
    • VantemioPassport;
    • VantemioWalletRegistry;
    • PlatformVault and PlatformVaultRegistry.
  2. Sign-in and wallet links:
    • Sign-In with Ethereum;
    • smart-wallet signatures (ERC-1271 and ERC-6492);
    • links to wallets in other networks;
    • the Passport.
  3. Swap and bridge:
    • quotes and the receiver of the coins;
    • the 0.4% service fee;
    • status tracking and request limits.
  4. Donations:
    • every supported network and coin;
    • swapping during a donation and the "stablecoins only" setting;
    • recording and confirming a donation;
    • chain reorganisations;
    • the background checker (keeper).
  5. Streamer tools:
    • donation QR codes: all of a streamer's networks; one network with or without one coin; the streamer's own set of several networks, each with all coins or chosen coins. Every kind has a "stablecoins only" option;
    • the OBS alert overlay;
    • message moderation and the filter;
    • the minimum donation;
    • wallet confirmation of sensitive changes (step-up);
    • the security log.
  6. Server:
    • authorisation and cross-site request protection;
    • request limits and input handling;
    • data exposure and module load-time behaviour on the edge runtime.

Method

  • Code review. We read the source of every component in scope and compared it with the platform's written architecture.
  • Tests for fixes. Each finding was proven by an automated test that fails on the code before the fix and passes after it.
  • Full test suite. After the fixes, the full suite ran with 0 failures. Last full run, after all run-3 fixes and the QR change: 3,692 tests, 0 failed (the rest skipped).
  • Contract checks:
    • static analysis;
    • unit and fuzz tests;
    • stateful invariant testing with three independent fuzzers.
  • Live checks. Run on the live site only as an ordinary visitor (read-only).
  • Test networks only. Every on-chain test used a test network with a test wallet. No mainnet transactions were made, and no real funds were used.
  • No load testing. There was no load or denial-of-service testing of the live site.

Tools and versions

All tools were installed from their official sources, and their published checksums were verified.

ToolVersionUsed for
Slither0.11.6Solidity static analysis
solc0.8.30compiler (the deployment settings: optimizer 200 runs, via-IR)
Foundry (forge)1.9.0, forge-std 1.9.7unit, fuzz and invariant tests
Echidna2.4.0property fuzzing
Medusa1.5.1property fuzzing
Semgrep1.185.0JavaScript and Solidity pattern rules
Aderyn0.6.8Solidity static analysis. Its maker has retired the tool and withdrawn the ready-made downloads, so we built it on macOS from the official release source (checksum verified)

Contract results

Audited source (SHA-256)

e753ee99…a2c804  DonationSplitterV2.sol
eab6a488…ec4ecd  DonationSplitterV2Tron.sol
7fd2efb0…5193a8  IDonationSplitterV2.sol
5f31592c…9e5ec0  VantemioPassport.sol
f84c02b5…ad8177  VantemioWalletRegistry.sol
e50ef072…02f2    PlatformVault.sol
8b915253…a570    PlatformVaultRegistry.sol

Full hashes are in the audit records.

Live contracts match the audited source

On 17 November 2026 we read the deployed code of these contracts from two public nodes per network, read-only. Both nodes returned the same code:

  • the donation splitter 0xC6ecafD48f8F1Aa8003FCA13e3738CbAfA097423 on Base, Arbitrum, Ethereum and Optimism;
  • the Passport 0x471c955fF085F3384a8a2d4f707D703374c84fc6 on Base;
  • the wallet registry 0xE15C9AE8a7B73097fD98bB3131Eaa104d5d0946E on Base.

We compared that code with our own build of the audited source, made with the same compiler settings. All six contracts match byte for byte. Two parts were set aside before comparing:

  • the compiler's metadata tail;
  • the constructor-set (immutable) values, which we checked separately.

The code hashes also match the deployment records.

Donation splitter: properties tested

The fuzzers (Echidna: 100,080 calls; Medusa: about 218,000 calls; Foundry invariants: 16,384 calls) found no sequence of donations that breaks any of these properties:

  • The contract never keeps coins or tokens after a donation.
  • The creator's part plus the fee always equals the donation. The fee is exactly the amount × rate ÷ 10,000, rounded down, for each rate: 1%, 0.5% and 0%.
  • A donation ID or donor nonce is paid at most once.
  • Nobody but the donor can move the donor's tokens, even when the donor has approved the contract.

Unit and fuzz tests

31 of 31 tests pass. Fuzz tests ran 5,000 times each.

The tests also cover:

  • refusal of tokens that take a fee on transfer;
  • refusal of tokens that report a transfer they did not make;
  • blocking of re-entry through the recipient and through a token callback;
  • donations still working, with an exact split, when someone has forced coins onto the splitter without a call.

On the Passport, wallet registry and vault, they cover:

  • the soulbound lock;
  • vouchers that bind to their sender and cannot be replayed;
  • the issuing rate limit;
  • revocation, which removes old vouchers, delegates and pending recovery;
  • controller change by signature;
  • delayed recovery with the holder's veto;
  • wallet link replay and purpose checks;
  • vault access control and exact accounting.

Static analysis

Slither reported 35 items. We reviewed each one, and none was an exploitable defect. The two "high" categories are explained in the code:

  • the "arbitrary send" item: the contract requires that the donor is the caller;
  • the "reentrancy" items: the TRON variant uses a re-entry lock and deliberate balance checks.

Aderyn reported 3 high and 13 low items. We reviewed each one, and none was an exploitable defect:

  • "state change after external call": every flagged call only reads data from an approved token or from the platform's Passport, under a re-entry lock or owner-only access;
  • "strict equality on contract balance": the splitter's check is relative to the balance before the donation. Coins forced onto the contract do not block donations; a test confirms this;
  • "contract name reused": the same sanctions-list interface is declared in two files;
  • the low items are owner powers (described in C-5 and C-6 below), the TRON variant's deliberate handling of tokens that return false, and style or gas notes.

Semgrep found no issues in 26 server and client files. On the contracts it gave gas-saving suggestions only.

Findings

Severity follows impact and likelihood: high, medium, low, info. The status is "fixed" only when a test that failed before the fix passes after it and the full suite is green.

Audit run-2 (15–17 November 2026)

Totals. Run-2 found 125 issues. All but one are fixed and released; that one is an informational header on static pages, described below.

AreaIssues
Swap and bridgeabout 35
Donation cardabout 15
Donations serverabout 10
Sign-in and Passportabout 10
Otherthe rest

Most serious, all fixed:

  • Swap, Solana:
    • A manipulated route quote could include extra transfers out of the wallet. The server and the page now check every outgoing transfer, account by account, in a simulation.
    • A fake "rent deposit" line could widen the allowed outflow. These lines are no longer trusted.
  • Swap, EVM or TRON to Solana or TRON: for these routes the receiver is now read from its exact place in the route data.
  • Swap, other networks: on 30 networks the route provider uses different contract addresses. Each network's checks are now bound to that network's own contracts.
  • Donations:
    • An EVM donation was credited only if the donor kept the page open. A background checker now finds settled donations on-chain.
    • A replaced Bitcoin transaction could lock a donation. The verifier and the checker now accept the replacement.
  • Donation card: four issues in the "other coin" and wallet-boundary flows.

Still open: the static pages send Access-Control-Allow-Origin: *. These pages hold no user data, and no request can carry the user's cookies, so this exposes nothing.

Audit run-3 (17 November 2026)

STEP-UP-1 — medium — fixed, live in API 95. Sensitive changes to a donation page need a signature from the page's main wallet over a readable message. That message cut the change details at 600 characters, while the server checked the whole request.

  • Attack. A long padding field could push the actual new payout wallet out of the visible text.
  • Precondition. The attacker needs a stolen session and must trick the owner into signing. Then the owner could approve a payout wallet they never saw.
  • Fix. The message now shows the details in full. The payout endpoint also refuses requests with conflicting or unknown fields.
  • Confirmed by:
    • a test that reproduced the redirect before the fix;
    • the same test passing after the fix;
    • the full suite green.

STEP-UP-2 — low — fixed, live in API 95. Four overlay settings could be loosened without the wallet signature:

  • turning messages back on;
  • lowering or removing a per-coin threshold;
  • enabling test mode, where free test-network coins could flood the stream;
  • unblocking a donor.

These changes now need the signature, like the other loosening changes. Tightening a setting still takes one click.

  • Confirmed by:
    • a test that failed before the fix;
    • the same test passing after the fix;
    • the full suite green.

Contract observations (info, no change needed):

  • C-1. Donation IDs are tracked per donor. The server credits a donation only if the transaction sender is the donor who requested it.
  • C-2. The on-chain wallet registry cannot link two kinds of wallet:
    • an EIP-7702 account whose delegate does not implement ERC-1271;
    • a smart wallet that is not deployed yet.
  • C-3. Changing the Passport controller by signature works for key wallets only. Smart-wallet holders use the delayed recovery.
  • C-4. Coins sent to the vault without a call (for example, by self-destruct) are not counted as income, although they can be withdrawn.
  • C-5. The platform owner (a multisig) is trusted to:
    • issue passports;
    • set roles and delegates;
    • start recovery, which waits 48 hours and which the holder can cancel.
  • C-6. Who controls the splitter differs by network.
    • On Base, the splitter's owner and fee recipient is a multisig contract.
    • On Arbitrum, Ethereum and Optimism, both are a single-key account. The fee recipient cannot be changed: it is set when the contract is deployed.
    • In every network, the donor's coins go to the creator in the same transaction. The owner cannot take donors' or creators' funds.
    • The owner can pause donations and change settings: the accepted coins, the fee rates of 1% or less, and the sanctions list source. Until ownership on these three networks is moved to the multisig, this single key is trusted with these rights. Its fee recipient stays as deployed.

Release incident (not a vulnerability)

On 17 November 2026, from about 06:02 to 06:05 UTC, two website releases (137 and 138) went out from a shared working folder that held unfinished payment markup. For about three minutes the donation card pointed to a script that the API did not serve yet. The card failed to load, so no payment could be started in that time and no funds were affected. API release 96 closed the gap. Each website release is now checked against the scripts the API actually serves before it goes out.

QR codes added during the audit

The audit noted that QR codes covered only one network with up to one coin, or all of a streamer's networks. The platform then added the streamer's own set (API 96). On the server, every network in the set must be one the streamer receives in, and every coin must be offered in that network. In "stablecoins only" mode, only the network's stablecoin is accepted. We reviewed the parser and checked the live site as a visitor: malformed sets, and sets combined with a single network, are refused.

What was not checked

  • Third-party code. The route provider's contracts and the networks' own bridges were treated as outside parties. We check what their quotes and transactions do; we did not audit their code.
  • Wallet apps. Wallet apps and browser extensions are outside scope.
  • Live-site testing. There was no load or denial-of-service testing of the live site, and no testing of mainnet funds.
  • Formal verification. The contracts were not formally verified.

Known limits

  • A lost page wallet key. Sensitive changes to a donation page need a signature from the page's main wallet: payouts, the minimum, the payout coin, the OBS link and loosening the filter. If that wallet's key is lost, the streamer cannot make these changes alone. There is no support procedure for this yet. The streamer can create a new donation page with a new wallet.
  • TRON donations and the keeper. The background checker does not search for TRON donations. A TRON donation is confirmed when the donor's page or a later report submits it.
  • Replaced Bitcoin transactions. A replaced Bitcoin transaction that nobody reports is found only when the donor's page or a new report submits it.
  • Smart-wallet signatures. These are checked on Base by two independent nodes. A node that wrongly answers "invalid" can delay a sign-in, but it cannot let anyone in.
  • This is not an independent audit. Our own AI auditor did this work. An independent audit firm may find issues we missed.

Evidence

The test files, tool outputs and before/after results are kept in Vantemio's internal audit records. They are available on request.

We use only necessary cookies: Passport sign-in and Cloudflare protection. No ad trackers; Cloudflare statistics work without cookies. Privacy policy

AcceptNecessary only