ForkFree

Prompts / Agents and automation

Safe Step-by-Step Device Cleanup Agent

You get a cautious, step-by-step clean-up plan where nothing changes without your approval and everything can be undone. It works because it sets the risks, protected items and approval rules before the AI acts.

Worked in our testThe reply starts with a plain-English risk warning, says it can make mistakes, ends with a confirmation question, and stops there. That is correct, because the audit and first change should only come after the user replies. It also covers the protected list, the undo approach and the single restart.

The prompt

Act as a cautious IT technician helping a non-technical owner tidy up a slow device.

Task: Help me speed up my {device_model} by turning off unneeded software, without risking it breaking.

Steps:
1. Before touching anything, explain the risks in plain words, mention that you can make mistakes, and ask me to confirm before going on.
2. Audit first: list every item in plain English under "safe to turn off", "needed by the device" and "I use this". Treat these as ones I use: {apps_i_use}.
3. Never touch anything on this protected list: {protected_items}. If unsure, leave it on.
4. Ask me before each change. Disable rather than delete, and keep a running undo list showing how to reverse every change.
5. Finish with one restart, check it boots normally, and compare performance before and after.

Output:
Start with the risk warning and a confirmation question. After my reply, give the audit as three headed lists, then propose the first change only, with its undo command.

Change the parts in curly braces. Use it when you hand an AI agent control of a device or system where mistakes are costly and every change must be reversible.

What happened when we ran it

We sent

Act as a cautious IT technician helping a non-technical owner tidy up a slow device.

Task: Help me speed up my Sony Bravia KD-55X80J Android TV, about five years old, which takes 40 seconds to open apps and often freezes on the home screen by turning off unneeded software, without risking it breaking.

Steps:
1. Before touching anything, explain the risks in plain words, mention that you can make mistakes, and ask me to confirm before going on.
2. Audit first: list every item in plain English under "safe to turn off", "needed by the device" and "I use this". Treat these as ones I use: Netflix, BBC iPlayer, YouTube, Disney+, Amazon Prime Video and Spotify.
3. Never touch anything on this protected list: Google Play services, Google Play Store, Settings, System UI, the keyboard, Bluetooth, HDMI inputs, and the TV's input and tuner framework. If unsure, leave it on.
4. Ask me before each change. Disable rather than delete, and keep a running undo list showing how to reverse every change.
5. Finish with one restart, check it boots normally, and compare performance before and after.

Output:
Start with the risk warning and a confirmation question. After my reply, give the audit as three headed lists, then propose the first change only, with its undo command.

The model replied

# Before we start: the risks I'll be working like a careful technician, so here is what you should know first. **What we're doing:** switching off (not deleting) apps and background services on your Sony Bravia KD-55X80J that you don't use, so the TV has more memory and processor time for the things you do use. **The risks, in plain words:** - **I can make mistakes.** I can't see your TV, and Sony's software varies between versions. A name that looks harmless to me might be something the TV quietly relies on. - **Turning off the wrong item could cause problems.** Examples are a broken home screen, a remote that stops working, missing channels, or Wi-Fi that won't connect. These are rare if we stay careful, but they're possible. - **Disabling is reversible.** Every change can be undone, and I'll keep a running undo list. A **factory reset** is the last resort if something goes badly wrong, and it would erase your logins and settings. - **Speed gains may be modest.** A five-year-old TV with limited memory may only get a bit faster. Freezing can also come from low storage, an old software version, or failing hardware, which switching things off won't fix. - **Software updates may switch things back on.** Sony or Google updates can re-enable apps, so we might need to repeat a change later. **How I'll keep it safe:** - I'll audit first and change nothing until you approve each step. - I will not touch Google Play services, Google Play Store, Settings, System UI, the keyboard, Bluetooth, HDMI inputs, or the TV's input and tuner framework. If I'm unsure about an item, I leave it on. - One change at a time, each with its undo command. - One restart at the end, then a check that it boots normally and a before/after speed comparison. **What I need from you:** - Your TV on the same Wi-Fi network as a computer, with Developer options and network (ADB) debugging switched on. I'll guide you through that if needed. - Your TV remote handy, in case we need to undo something. ## Do you understand these risks and want me to continue? Please reply **yes** to proceed, or tell me if you'd like to change anything first.

Run on Claude Sonnet 5.5, 01 Nov 2026. Automatic check: 5 out of 5.

Mentioned for

Claude Code, Codex

Where it came from

Found on GitHub, shared by Ansub: Ansub/tv-detox.

This is our own version of the idea, written and tested by us. The original may differ.

More agents and automation prompts

Agents and automation Worked in our test
Propose, Probe, Compose Prompt Refiner

You get a tested, improved set of instructions plus short lessons you can reuse next time. It works because each edit is judged by comparing what it actually changes in the output, rather than trusting your first impression.

Found on GitHub, shared by google-research
Agents and automation Struggled in our test
Turn a Recurring Task Into Instructions

You get a clean, reusable instruction file for a task you repeat, with the role, steps, decision rules and output format kept apart. Splitting who, how and what makes the instructions easier to test, reuse and improve.

Found on GitHub, shared by alirezarezvani
Agents and automation Struggled in our test
Turn a Successful Run Into a Skill

You get a short, reusable skill file that records the exact route that worked, without the failed attempts. It works because the agent captures the path while the session is still fresh, and skills load only when needed.

Found on Web, shared by @hiper2d (Aliaksei Zelianouski)