Summary
I used CleanYourTV with Claude Code to debloat my Android TV over Wi-Fi. The agent backed up the package list, disabled the sponsored and telemetry packages, built a small launcher, and made it Home. The TV felt faster and easier to use afterward. Nothing was uninstalled, no root was needed, and the repo left a one-line restore path. The setup worked, but the build prerequisites were the part that needed attention before the run. This is what I did, what changed, and what I would check before doing it again.
Introduction
My Android TV worked. It just felt crowded and slow. The home screen had a sponsored row, recommendation rails I never used, and a handful of preloaded apps that could not be removed from the normal settings. Moving around the interface felt like waiting for the TV to catch up.
I did not want to root the TV. Rooting changes the support story, and a mistake can be hard to undo. I also did not want to buy another streaming box just to get a quieter home screen.
CleanYourTV takes a third path. It uses the Android debug bridge that is already on the TV, the same adb developers use on phones. It connects over Wi-Fi, disables packages with pm disable-user, and installs a small native launcher. Nothing is uninstalled.
The project is at Gathin23/cleanyourtv. The site is cleanyour.tv. It supports Android TV and Google TV, with partial support for Fire TV. Samsung Tizen, LG webOS, Roku, Vizio, and Hisense VIDAA do not expose ADB, so the debloat half does not apply there.
I tried the repository's Claude Code workflow on a normal Android TV. This is the account of that run.
Why I tried it
The TV is a computer with a remote. The manufacturer controls the home screen, the app list, the recommendations, and the update channel. On a set with limited memory, that control costs time before you open anything.
The usual advice is to disable a few apps by hand. That works until the next update puts them back, or until you hit a package that the settings app will not let you touch. A list plus a repeatable script is easier to reason about.
What convinced me was the restore story. The repository disables packages instead of uninstalling them, writes a record before it changes anything, and ships a restore script. That is a different risk profile from a root exploit or a custom firmware image.
I also wanted to see whether a coding agent could handle a device workflow without hand-holding. The repository prompt is written for exactly that.
What I did
The quick start assumes a computer on the same Wi-Fi as the TV, with Python 3, a JDK from 17 through 22, and Android SDK platform 34 with build-tools 34. The agent needs those to build the launcher.
On the TV, I opened Settings, went to Device Preferences, then About, and pressed OK on Build seven times. I went back to Developer options and turned on USB debugging. The TV also had a network debugging toggle, so I turned that on too and kept the TV awake.
On the computer:
git clone https://github.com/Gathin23/cleanyourtv
cd cleanyourtv
I opened that folder in Claude Code, pasted PROMPT.md, and filled in the TV details. The agent asked for the model, then walked through a checklist: which apps I use, the dock size and order, whether to keep the voice assistant, whether to keep Chromecast, and whether to keep automatic updates.
I accepted the "Allow USB debugging?" dialog on the TV. The agent then took over.
The sequence it followed was the one in the repository:
- Probe the TV and list the installed packages.
- Write a backup record before changing anything.
- Disable the packages in the chosen list with
pm disable-user. - Build the launcher.
- Install it, make it Home, and check the result with screenshots pulled from the TV.
Cloning the repository mattered. The agent reused tools/disable.sh and launcher/build-and-install.sh instead of writing its own scripts. Those files contain the backup, the Home guard, and the restore record. A prompt alone would not have carried them.
What actually happened
The run worked. The TV booted into the new launcher, and the sponsored row and the recommendation rails were gone. The apps I said I used were still there and still opened. Nothing had been uninstalled, so the package list was shorter in behavior, not in storage.
The change I noticed most was speed. The home screen responded faster, and moving between apps no longer felt like waiting for the set to catch up. I cannot credit one package for that. The ad services, recommendation rows, and preloads all do work before you press anything, and removing them cut that work. The TV went from a sluggish mess to something I can use without thinking about it.
The launcher is small and plain. It has a dock, a grid that slides up on Down, and buttons for inputs, network, and settings. The focused app gets a top shelf with a glow taken from its artwork. It is closer to an Apple TV home screen than to the stock Google TV layout.
The build was the slowest part. The agent had to build a release APK, install it, and confirm that Home landed on it. The repository estimates about fifteen minutes for the whole run. The estimate depends on the computer having the right JDK and Android SDK before the agent starts. If those are missing, the agent spends its time installing tools instead of changing the TV.
I did not need the restore path. I kept the new launcher as Home.
The parts that made it safe
Three details did most of the work.
The first is that pm disable-user is reversible. The package stays on the system partition, so a restore can turn it back on. That is why the project can promise a one-line undo without a factory reset.
The second is the Home guard. The scripts ask the TV which package owns Home. They only disable a stock launcher they recognise, and they skip that entry if no other app can answer Home yet. The TV is never left without a home screen. I did not have to think about that ordering, because the scripts do.
The third is the record. Before disable.sh changes anything, it writes the state it needs to reverse the change to ~/.cleanyourtv/. Restore replays it, so a package I had disabled myself stays disabled, and Home returns to the package that owned it.
TV_HOST=192.168.1.50 tools/restore.sh examples/<your-tv>/debloat.txt
If I had wanted the stock launcher back on Google TV, the command takes that package as an argument. I did not need it.
What I would watch next
The change lives outside the manufacturer's software, so a firmware update can undo it. An update installs over the cleaned system, the listed packages return, Home may go back to the stock launcher, and new preloads can appear that no list mentions.
The repository has tools/check.py for that. It compares the TV with the list you applied, and --fix puts the disabled packages back.
TV_HOST=192.168.1.50 tools/venv/bin/python -I tools/check.py examples/<your-tv>/debloat.txt
I would run that after the next update, not before.
The other thing I would watch is the ADB port. Network ADB is a remote shell on the TV. I turned USB debugging off when the run finished. The repository includes tools/adb-closed.py, which checks that the port refuses connections. A pass means the TV is closed. A failure means it is still open.
The platform limits are real. Fire TV can run the debloat half but will not hand Home to a third-party launcher. Some China-market TCL sets refuse to install any launcher. Samsung, LG, Roku, Vizio, and Hisense VIDAA do not expose ADB at all.
Best practices and common pitfalls
Best practices
Install the prerequisites before you start the agent. Python 3, a JDK from 17 through 22, and Android SDK platform 34 with build-tools 34. The run is smoother when the agent does not have to fix the toolchain first.
Clone the repository and let the agent use its scripts. The backup, Home guard, and restore record are in those files.
Keep the TV awake and on the same network for the whole run. A TV that sleeps during the build or the install can leave the process halfway through.
Save the debloat list you applied. It is the baseline for restore and for the next firmware check.
Turn debugging off after the job and verify the port is closed. Run the check, do not assume.
Common pitfalls
Do not disable the stock launcher before a replacement answers Home. The install script orders this for you, but a hand-run workflow can get it wrong.
Do not disable the Google TV setup wizard without checking the Wi-Fi state. On some sets the saved network belongs to the wizard process.
Do not treat the fifteen-minute estimate as a promise. A missing JDK or an Android SDK mismatch will add time.
Do not use a list for a different model without review. Package names differ between brands and firmware versions.
Do not leave network ADB enabled for convenience. It is a shell, not a feature.
Key takeaways and conclusion
Key takeaways
- CleanYourTV worked on a normal Android TV without root, and nothing was uninstalled.
- The set felt faster and more usable after the ad and recommendation services were disabled.
- The reversible operation and the restore record are the reason the risk is acceptable.
- The Home guard is the one detail I would copy into any device automation.
- The next real task is the firmware check, not the first run.
Conclusion
I went in expecting a demo and got a faster, usable TV. The interesting part is not the launcher, though the launcher is fine. It is the restraint. The tool disables instead of uninstalling, records state before changing it, refuses to remove the only home app, and gives you a line to undo the change.
That pattern applies beyond TVs. When an agent is going to change a device you depend on, give it a reversible operation, a guard against the one change that can strand the device, and a record it can replay.
I kept the new launcher. I will run the firmware check after the next update, and I will keep the ADB port closed until then.
References
- CleanYourTV repository and README. https://github.com/Gathin23/cleanyourtv
- CleanYourTV site and supported TV list. https://cleanyour.tv
Powered by Mole. AI Agent for SEO & GEO
Top comments (0)