A Claude skill is a folder of written instructions. The main file is called SKILL.md, and a few reference files usually sit beside it. Until a request matches, Claude keeps only the skill's name and its one-line description in mind. Then it opens the folder and follows the method inside, reading a reference file only when a step calls for it. That is where a skill beats a long prompt pasted into the chat. The method arrives the same way in every chat, and it takes up little room until it is needed, which leaves space for your own photos and notes. You install it once and Claude can use it in any new conversation.
On claude.ai and in the desktop app the steps are the same. Under Settings, then Capabilities, check that Code execution and file creation is on. Open Customize in the left sidebar, choose Skills, click the + button, then Create skill and Upload a skill, and pick the zip exactly as it downloaded. Anthropic's help centre says every plan can do this, Free included, though an owner can turn it off on Team and Enterprise. In Claude Code there is nothing to upload. You unzip the file into ~/.claude/skills/ so that the main file sits at, for example, ~/.claude/skills/product-photo-shoot/SKILL.md. Do all of this on a computer, because Anthropic's pages do not cover custom skills in the phone apps.
Start each job in a new chat and name the skill in your first message, so Claude does not have to guess. Attach what the skill works from: clear photos of the product, your logo, a page of brand guidelines, a few past posts, the brief you were given. Answer its questions one at a time. If it drifts, tell it which stage you want redone and what to change, or ask it to reread the skill and pick up from there. When a newer version comes out, download the new zip. On claude.ai, upload it and accept the offer to replace the old one, or delete the old copy first. In Claude Code, delete the old folder and unzip the new one in its place.
The prompts
01Install a skill in Claude CodeTerminal mkdir -p ~/.claude/skillsunzip ~/Downloads/product-photo-shoot.zip -d ~/.claude/skills/# Then type /skills in Claude Code and look for product-photo-shootSwap product-photo-shoot for your skill's id, which is the zip's file name without .zip. An open session finds the skill without a restart, but if this is your first skill, type /reload-skills once. If Safari has already unzipped the download, run mv ~/Downloads/product-photo-shoot ~/.claude/skills/ in place of the unzip line.
02A first message that starts any skillText Use the [SKILL NAME] skill. I am working on [THE JOB, FOR EXAMPLE A LAUNCH POST FOR OUR CANDLE]. I have attached [WHAT YOU ATTACHED, FOR EXAMPLE PRODUCT PHOTOS AND OUR LOGO]. It will be used on [WHERE IT WILL RUN] and it is for [WHO IT IS FOR]. Read the skill first, then ask me what you still need, one question at a time.Naming the skill is what Anthropic suggests when Claude is not picking one up. Asking it to read the skill first also lets you see that it did.
Checklist
Before your first job with a skill
- Code execution and file creation is on, under Settings, then Capabilities, and you are on a computer.Skills run in Claude's code environment, so with this off the Skills page is missing or greyed out. On Team and Enterprise plans an owner sets it under Organization settings, Plugins & skills, in the Policy tab.
- You uploaded the zip exactly as it downloaded.claude.ai expects one folder at the top of the zip, named the same as the skill. Unzipping or renaming it is the usual reason an upload fails. Safari on a Mac unzips downloads unless you turn that off in its settings.
- The switch next to the skill is on, in Customize, then Skills.A skill that is switched off is not available to Claude, however clearly you ask for it.
- Your first message names the skill and says what the job is.Claude can pick a skill up from the task alone, but naming it removes the guesswork. In Claude Code you can type a slash and the skill's id instead.
- You attached the material the skill works from.Product photos, a logo, brand guidelines and past posts give it real facts to work with. Without them it has to stop and ask, or leave gaps for you to fill.
- Claude showed a step where it read the skill before it began.If you cannot see it open the skill, ask it to read the skill first and start again. In Claude Code, /skills lists every skill it found.
- When you install a newer version, the old copy goes.Two copies with one name make it unclear which one Claude reads. On claude.ai, accept the offer to replace it or delete the old one first. In Claude Code, delete the old folder before you unzip the new one.
What to do it with
- ClaudeFree tier, paid above it
A language model that writes, edits, reads long documents and runs tools, including agents that work inside a codebase.
Bad at It is agreeable by default, so it will improve a bad idea instead of telling you it is bad. Ask it to argue against you and the quality of the answer changes.
- Claude CodeSubscription
An agent that reads a whole repository, edits files, runs the tests and explains what it changed.
Bad at It is fast enough to make a large mistake quickly. Work on a branch, read the diff, and never let it push. Watch the starting mode: on Pro, Max and Team plans it opens in auto mode, which Anthropic's own list says may read a .env file and send the key in it to the API that key belongs to. Switch it off.
Why it matters
A long prompt is a brief you rebuild by hand every time, and small parts of it go missing between chats. A skill keeps the method in one file, so Claude asks the same questions and runs the same checks on the tenth job as it did on the first. When the method improves, you swap in a new file and carry on with the work you already have.
- Type
- technique
- Period
- ongoing