Yes. The Agent Skills format is open and free. A skill is a folder with a SKILL.md file inside it, plus optional scripts and reference files, and Anthropic does not charge a separate fee to create one or run one. The standard lives at agentskills.io, and Anthropic's own skill library ships under Apache 2.0 in the public skills repo. You write a markdown file, drop it in a directory, and Claude can use it. No license key. No marketplace toll.
So when people ask "are claude skills free," the format is genuinely the free part. What carries a price is context, and that price is small but real. I want to draw the line clearly here, because most people fold the two together and end up confused about what they are actually paying for.
Are Claude skills free to create and install?
The format is free. If you are still fuzzy on what a skill even is, the primer on what a Claude skill is covers the basics. On cost, here is the short version. Making a skill means authoring a SKILL.md file with YAML frontmatter (a name and a description) and a body of instructions. That is the whole requirement, per Anthropic's docs. You can keep it on your filesystem in Claude Code, upload it through the API, or add it in claude.ai settings, and nothing in that loop bills you for the skill itself.
What you pay for is whatever your Claude plan or API usage already costs. A skill adds no surcharge on top. It runs inside the same token accounting as everything else Claude reads and writes.
Do Claude skills use tokens?
They do, and this is the part people miss when they ask about claude skills cost.
Skills load through progressive disclosure, in three stages. At startup, only the metadata loads. Each skill's name and description go into the system prompt so Claude knows the skill exists and when to reach for it. Anthropic's docs put that at roughly 100 tokens per skill, which is the always-on tax, and it is tiny. The full SKILL.md body only enters context when your request actually matches the skill's description, and even then it is meant to stay under about 5,000 tokens. Bundled reference files and scripts sit on the filesystem at zero token cost until Claude reads or runs them, which is the whole point of the design.
So the honest answer to "do claude skills use tokens" goes like this. The descriptions always cost a little. The body costs more, but only when triggered. The bulky stuff costs nothing until something needs it. A well-built skill you never trigger in a session runs you a few dozen tokens for its description and nothing else.
This matters because context is finite. Every token a skill description occupies is a token your actual conversation cannot use.
Is there a claude skills limit on how many you can have?
There is no hard cap on the number of skills you can install. You will not hit a wall that says "skill 51 rejected." There is a practical limit, though, and it comes straight from that always-on description budget.
Here is the mechanism, and the deeper write-up on the skill listing budget goes further than I will. In Claude Code, every skill's name is always listed, and the descriptions get loaded so Claude can match your request to the right skill. Those descriptions share a budget that scales at roughly 1% of the model's context window. Pile on enough skills and that budget overflows. When it does, Claude Code starts shortening descriptions, dropping the full text of the skills you invoke least first. A shortened description can strip the exact keyword Claude needed to pick the right skill, so the skill quietly stops triggering. Run /doctor and it will tell you whether your budget is overflowing.
That is the real answer to "how many skills can Claude have." The cap is not on the count. It is on the budget. Two hundred lean skills with sharp twelve-word descriptions can fit where forty bloated ones with paragraph-long descriptions will not.
What this means in practice
Keep skills few and lean. A skill earns its place by getting triggered, and one that never fires still spends description budget for the privilege of sitting there. Write descriptions that stay short and keyword-dense so Claude matches them reliably without eating the budget. Push long reference material into bundled files that load on demand, rather than stuffing it into the SKILL.md body. If a skill has not fired in months, delete it.
A free format is what makes this discipline possible. You can author, test, throw away, and rewrite skills without anyone charging you per attempt. Building a skill with Knack turns a short interview into a clean SKILL.md folder, and you pay for the build rather than the standard. The standard stays the same open Anthropic format that runs in Claude Code, the Claude apps, Codex, Cursor, and Gemini CLI, and you can take it anywhere for free.
Free to own, cheap to carry, costly only if you hoard. Build five good skills before you reach for fifty mediocre ones.