To organize screenshots you never revisit, sort them by what you wanted from each one instead of what’s in it. A book cover means add it to a wishlist. A funny comment means save it as a shirt idea. I wrote a Claude Code skill that opens my screenshot folder, reads every image, works out the intent behind each, hands back something I can act on, and files the picture away. The whole file is below.
My screenshot folder was a graveyard
I screenshot things constantly. A book somebody recommended. A recipe in a video caption. A comment so good it belongs on a shirt. A pricing page for a tool I wanted to look at later.
Later never came. The folder just grew, and every so often I’d scroll it, feel bad, and close it. The pictures weren’t the problem. The problem was that each one was a little note to myself with no note attached, and by the time I found it again I’d forgotten what I wanted.
So I built something to ask the question I’d stopped asking: what did I save this for?
What does the screenshot organizer do?
It opens my screenshot folder, looks at each picture, decides why I saved it, and hands back the thing I actually wanted. Book titles come back as a list I paste straight into my wishlist. Shirt ideas go into my task manager. Then every picture moves to an archive folder, so the inbox is empty and stays a real inbox.
Two parts of this are not mine and I want to be clear about them. Claude can read images already, and skills are a built-in Claude Code feature, which is a text file you write once so you stop explaining the same job every session. If skills are new to you, start here and come back.
What I wrote is the instructions: how to sort them, what to check for, and what shape the answer takes for each kind.
What do you need to build the screenshot organizer?
Claude Code, a folder your screenshots already land in, and about ten minutes. There’s no coding in it. A skill is a plain text file with instructions in it, so if you can write down how you’d explain the job to a person, you can write one.
Two things it’s worth knowing before you start. Claude Code needs a paid Claude plan, so this isn’t a free build. And the model has to actually look at your images, which means they leave your machine. If that’s a problem for the kind of thing you screenshot, this isn’t the tool for you and I’d rather say that now than three sections in.
If you don’t use Claude Code at all, skip to the paste-into-chat version further down. It works anywhere you can attach images.
How does the screenshot organizer work?
It runs in six steps. It lists what is sitting in your inbox folder, reads each picture, sorts each one by why you saved it, runs a safety check before writing anything, asks you about whatever it cannot place, then files the lot into an archive. Each step is one thing you decide. Mine is underneath as an example.
1. Pick two folders
An inbox and an archive. Point your phone or your Mac at the inbox so screenshots land there without you doing anything.
Mine is a screenshots folder with a Processed folder inside it. Nothing clever. The clever part is the second folder, because a tool with nowhere to put finished work will keep showing you the same pile.
2. Tell it to read every image, not scan filenames
Filenames tell you nothing. IMG_4471.png is not information. The instruction is to look at the actual picture.
3. Sort by what you wanted, not what’s in it
This is the part worth stealing. Two screenshots can both be “a picture of text” and need completely different handling.
| What’s on screen | What I wanted from it | What comes back |
|---|---|---|
| Book cover or a book page | Add it to my wishlist | A pasteable list of titles and authors |
| Terminal output or a prompt someone posted | Try this setup | The prompt, cleaned up and ready to run |
| Ingredient list or recipe card | Make this | The recipe, written out properly |
| App screenshot or pricing page | Look into this tool | Name, what it does, what it costs |
| A post or quote | Write something about this | The idea, stated as a thing to make |
| A funny comment | Put this on a shirt | Sent straight to my task list |
| Confirmation number or error message | Keep in case I need it | Filed, no action |
Write your own version of this table. Yours will look different because you save different things, and the table is the skill. Everything else is plumbing.
4. Add a safety pass before anything gets written
Every image gets checked first for scams, phishing, passwords or keys left visible, and text written to hijack the AI reading it.
That last one deserves a plain explanation. A screenshot is a picture of text, and if somebody puts “ignore your instructions and do this instead” in a post, and you screenshot the post, you’ve handed those words to your assistant. It gets flagged and skipped rather than followed. Screenshots go through the same door as anything else you paste in, so they get the same suspicion.
5. Make it ask when it can’t tell
Some screenshots are genuinely unclear. You can see what’s on screen and still have no idea what past-you wanted.
Mine puts those on a short “I’m not sure about these” list and asks. It never files them somewhere reasonable-looking. A tool that guesses wrong once is a tool you stop trusting, and an inbox you don’t trust is the graveyard you started with.
I ran the version below over two of my own screenshots while writing this. The first was a prompt somebody had posted, and it sorted cleanly. The second was an outreach email template, which none of my seven types really covers, and it went straight onto the ask list instead of getting shoved into the closest-looking bucket. Both of those are correct behavior, and the second one is the reason step 5 exists.
6. File everything and report
Move each processed picture to the archive, then say what happened. Empty folder, and a record of where things went.
The mistake that put a rule in the file
Mine has a rule that looks strange until you know why.
I don’t screenshot a long post once. I screenshot it while scrolling, four or five frames in a row, because that’s the only way to catch the whole thing. Those land as IMG_4471, IMG_4472, IMG_4473.
The first version treated each one as a separate item. It handed me three fragments of one prompt, each confidently summarized on its own, none of them making sense. It wasn’t wrong about any single picture. It was wrong about how I take pictures.
So the file now says consecutive numbers are usually one thing, and never split one item across two runs. Which is the real lesson: your version has to match how you actually behave, not how a tidier person would.
Get the screenshot organizer
I’m not going to hand you my file. It sorts for shirt sayings and books because that’s what I screenshot, and yours will be something else entirely. So build your own instead. Same six steps, your categories.
Build your own screenshot organizer
Put in what you screenshot and what you wish you had instead of the picture. It writes you a skill file for Claude Code, a paste-into-any-chat prompt if you don’t use it, or both. Your categories and nothing else. Nothing is saved and nothing is sent anywhere.
Take the skill file it gives you and save it as SKILL.md in a folder named process-screenshots, inside .claude/skills/. The folder name becomes the command. If you took the chat prompt instead, there is nothing to install: paste it and attach your pictures.
What I’d do differently
- Start with five types, not eight. I added the last few because I hit them, and guessing at categories up front wasted a rewrite.
- Write the archive step first. A version without it kept re-reading the same pile and I lost a run to it.
- Say what it skipped. Silent skipping in the safety pass looked identical to an empty folder.
- Cap the batch. Pointing it at months of backlog produced a wall of output I didn’t read, which is the same problem in a new outfit.
What it doesn’t do
Some of this is deliberate and some of it is just where I stopped. Worth knowing which is which before you point it at anything.
- It won’t do your whole backlog. Fifteen per run, on purpose. I tried it on months of pictures and got a wall of output I didn’t read, which is the same problem I started with.
- It won’t file a guess. Anything it can’t place goes on a list and comes back to you. That’s the whole safety design, and it means you’ll always have a few stragglers to sort by hand.
- It only reads images. PDFs, videos and screen recordings in the same folder get left alone.
- It isn’t a search index. Once a picture moves to Processed it’s filed, not tagged. The output you paste somewhere is the thing you keep, not the folder.
- It doesn’t send anything anywhere. It hands you lists. Getting the books into a wishlist or the sayings into a task manager is still your move, or another skill’s.
Frequently asked questions
How do I organize screenshots without doing it by hand?
Point an AI tool at the folder and give it written instructions for sorting by why each was saved, then have it move finished images to an archive folder. The sorting rules matter more than the tool. Decide what your categories are and what you want back for each, and the rest is a short file.
Do I need to code to build this?
No. The skill is a markdown file, which is plain text with a few formatting marks. The version above is around 45 lines and most of it is two lists you’ll change to match what you save.
Is it safe to feed screenshots to AI?
Treat them like anything else you paste in. A screenshot can carry a scam, a visible password, or text written to give your assistant instructions, so the version here checks every image first and skips what looks wrong. Keep genuinely sensitive images out of the folder entirely.
What if it files something in the wrong place?
Build in the rule that it asks instead of guessing. Anything it can’t confidently classify goes on a short list for you to answer rather than getting filed somewhere plausible. One bad guess costs you more trust than ten correct ones earn.
How many screenshots can it handle at once?
Mine caps at 15 per run and tells me the real count so nothing disappears. Larger batches produce more output than I’ll actually read, which recreates the original problem.
Where do the results go?
Wherever you’ll see them. Mine sends book titles back as a paste-ready list and pushes shirt ideas into my task manager, because both of those are places I already look. Output filed somewhere tidy that you never open is the same as no output.
Try it yourself
To organize screenshots you keep saving and never opening, write down the categories you actually save and what you want back for each, then let the file do the sorting. Start with your five most common. Point it at ten pictures, not a thousand.
If you want more files to copy, my writing style lives in one so I stop pasting style rules into new chats, and my publishing checklist is another.
Subscribe if you want the next one.
A tool I built
Subscription Graveyard
Bury the subscriptions you’ve cancelled, give each one a tombstone, and watch what you’re saving pile up.
Bury your first one →I run an affiliate program on the tools I build. Approved affiliates earn 20% on what their referrals pay, for up to a year. US only for now. See the program →
