Scanned, and shown as is.
Every download is run through VirusTotal, which checks it against about 67 antivirus engines at once. The results are below, including the ones that are not flattering. You can click straight through to the reports and re-run them yourself.
Nothing here is hidden, including the bad bits.
What the scanners say
Scanned 23 September 2026, on exactly the files this site serves. The counts are live links to the full reports, where you can press Reanalyse yourself. A word on the numbers: the engines that flag these are heuristic and AI-based, and they disagree with themselves — the count drifts by one or two between scans of a file that has not changed. None of them name real malware.
One engine flagged it. The other 65 did not — including every large name on the list.
- Zillya —
Dropper.Agent.Win32.746397 - The identical signature on all four downloads — see below. It is matching the packaging, not anything in the program.
a9824ecc2d22defff19eadd19d234d998bbf60505b327c6f0ada990abe9ae931
One engine flagged it. The other 66 did not.
- Zillya —
Dropper.Agent.Win32.746397 - The identical signature on all four downloads — see below. It is matching the packaging, not anything in the program.
830fc0fa5b5e9f30338bcffbe96917b42c0e4dbb15aefea4a0bdfd7b5a7c9c43
One engine flagged it. The other 58 did not.
- Zillya —
Dropper.Agent.Win32.746397 - The identical signature on all four downloads — see below. It is matching the packaging, not anything in the program.
d5cf126310998a673be6b2488620b5ecfa698cfe0fa288639273df546e044b17
Five engines flagged it, the other 61 did not. It is consistently the highest of the four — see below for what is different about it.
- Kaspersky —
HEUR:Trojan.Python.Agent.gen - Trellix ENS —
Artemis!D543A8ED1888 - Elastic — Malicious (high confidence)
- SentinelOne — Static AI – Suspicious Archive
- Zillya —
Dropper.Agent.Win32.746397
f2748c650023a736e03807b08ffd447035270c0b976161d8eaf1cc37c53857a5
Why one says “trojan”
The short version: those engines are guessing from shape, not recognising known malware. The names they use say so themselves.
Read the detection name
Kaspersky’s verdict is HEUR:Trojan.Python.Agent.gen. Every part of that is a
hedge. HEUR: means heuristic — a rule of thumb, not a match.
.gen means generic, a whole category rather than one identified thing. And
Python is the actual finding: it noticed a packaged Python program.
SentinelOne says “Static AI – Suspicious Archive”. That is a model’s opinion about a zip file it has not seen before. None of these engines claim to have found a specific piece of malware, because they have not.
A packed Python app starts like a dropper
These tools are written in Python and bundled with PyInstaller, so each one ships a
copy of the Python runtime alongside the program. Starting it means unpacking bundled
code and running it — which is, step for step, what a malware dropper does. Zillya’s
Dropper.Agent is flagging that shape.
The same bundler is used by thousands of legitimate programs and by malware authors too, so engines that match on the bundler catch both. That is the whole story behind the word on the screen.
The files are new and unsigned
A code signing certificate is what lets Windows and antivirus engines tie a file to a named company. These downloads do not have one yet, so every new build arrives as a file nobody has seen, from nobody in particular. Several engines weight exactly that combination — new, unsigned, packed — when they have nothing else to go on.
This is the honest gap, and it is the one thing on this page we can actually fix. A certificate is on the list.
The tools touch what scanners watch
SealTweaks edits the registry and services. SealClicker listens for a hotkey and sends mouse clicks. Seal Utility locks the cursor and manages what starts with Windows. Those are the same capabilities a scanner looks for, because malware uses them too. A behaviour engine sees the capability, not the intention behind it.
The same engine flags all four, with the same name
This is the clearest evidence on the page, and it arrived by accident. Zillya reports
Dropper.Agent.Win32.746397 — the same signature, down to the number
— on every one of the four downloads.
Those four programs have nothing in common. One edits the registry, one repeats mouse clicks, one locks a cursor to a monitor, one saves videos. They share no code, no libraries and no purpose. They cannot all contain the same piece of malware.
What they do share is how they are packaged: each is a Python program bundled with PyInstaller, so each one unpacks itself when it starts. That is what the signature matches. It is a fingerprint of the packer, not of anything we wrote — and it would fire the same way on a great many entirely ordinary programs.
The one that would light up, does not
If something genuinely malicious were hiding in these downloads, the obvious suspect would be SealTweaks — it is the one that changes the most on your system, needs administrator rights, and rewrites registry keys and services. It comes back 1 out of 66, from a single engine, with a generic label about unpacking. Every large name on that list — Microsoft, Kaspersky, BitDefender, ESET, Avast, Sophos, McAfee — comes back clean on it.
Why Seal Utility is the one with the most detections
It scores five, where the other three score one. Watch it over time and it went 3, 5, 4, 5, 5 across five scans — on builds that were, in behaviour, the same program. That drift is the tell: these are AI and heuristic engines guessing from shape, and they change their minds between scans. A real malware signature does not come and go like that.
The one real change that did move it was the licence check we added, so the app can tell whether it has been paid for.
That check does three things engines weight heavily when they are guessing: it makes a request over the internet, it runs a small background thread that wakes up later, and it writes a file to disk. Each one is ordinary on its own. Together, inside a packed Python bundle that nothing has a reputation for yet, they add up to a shape two more engines decided they did not like.
VirusTotal settles on dropper.python as the summary label, filed under
“dropper” and “trojan”. Read literally that is a description of
the packaging: a Python program that unpacks itself. Trellix’s
Artemis!D543A8ED1888 follows the same pattern as the others — Artemis
is its cloud-reputation guess, and the hex on the end is a fragment of the file’s
own hash, which is another way of saying “we have not seen this file before”.
We could have made this number look better by leaving the licence check out and shipping the paid app unlocked. That is not a trade worth making, and hiding the change would be worse than the number.
Slimming the download helped for a scan — an unused cryptography library and a 21 MB maths library that nothing ever called were removed — and then a later scan ticked back up by one. That is the drift, not a regression. Nothing dangerous was added or removed; a heuristic engine simply changed its guess.
The honest limit: this will not reach zero, and here is why
We tried. The build now strips every scrap of package metadata, so the earlier claim that the archive “masquerades as” some library is gone at the root. The two free apps come back completely clean. And yet SealTweaks still shows a detection, and Seal Utility still shows a handful. That is not a loose end we are still chasing — it is the floor, and it is worth explaining plainly.
Two things keep these numbers off zero, and neither is a hidden threat.
The first is the packaging itself. A program bundled with PyInstaller unpacks itself when it starts, which is the same opening move a real dropper makes, so one or two heuristic engines flag the shape regardless of what the program does. Almost every packaged Python app on the internet carries a flag or two for this reason. The only real cure is a code-signing certificate, which ties each build to a verified publisher and lets engines trust it — something we may add later; it costs money and an identity check, and it changes nothing about what the app does.
The second is specific to SealTweaks, and it is simply true. VirusTotal’s code analysis notes that the tool “includes functionality to disable critical security controls such as Windows Defender, SmartScreen, Core Isolation Memory Integrity (VBS), and CPU vulnerability mitigations.” It can. Eleven of its tweaks do exactly those things. We are not going to dress up an accurate description of our own product. What we will say is how they work: every one is off until you turn it on, listed by name with its own explanation of what it changes and what it costs, and reversible from inside the app. They exist because turning off Memory Integrity to recover lost frame rate is a real, well-documented trade, and people were doing it by hand from forum posts with no way back. Whether it is worth it is your call, which is why it is a switch and not a default. An engine that flags “can disable Defender” is not malfunctioning — it is describing the tool, and we would rather you saw that here than felt it was hidden.
Don’t take our word
Everything above is checkable without asking us anything.
Check the file is the one we scanned
Open PowerShell in the folder you downloaded to and run
Get-FileHash .\SealTweaks.zip. The value it prints should match the SHA-256
on the card above. If it does not, the file changed somewhere between us and you, and
you should not run it.
Run the scan again yourself
Open any report linked above and press Reanalyse. You are not looking at a screenshot we made — it is the live report for that exact file, and it will show you today’s verdicts from every engine rather than ours.
Decide for yourself
If this still does not sit right with you, do not install it. That is a reasonable call, and we would rather say so than talk you past it. The tools change real settings on your computer, so it is worth being careful about what you run.