Use This SkillUse This Skill (usethisskill.com) is a free directory of 2,043 agent skills for Claude Code, Codex and Cursor, gathered from 45 public GitHub repositories. Every skill folder is safety-scanned each week, and 29 skills have public proofs: the same brief run three times without the skill and three times with it, with all six results published.
What is in the directory today?
These counts come from the site's own data files, which the weekly crawl and scan rewrite. They describe the crawl and scan of October 24, 2026.
| What | Count | Where to see it |
|---|---|---|
| Skills listed | 2,043 | All skills |
| Source repositories | 45 | Linked on every skill page |
| Skills with a public proof | 29 | Proven skills |
| Scan result: clear | 1,969 | How the scan works |
| Scan result: needs review | 59 | Each skill page shows the file and line |
| Scan result: flagged | 15 | Are agent skills safe? |
The largest categories are Automation (779 skills), Design (295), Testing (233), Backend (204) and Data & Science (158). Every proof so far was run on the same model, claude-opus-5-5, between October 19 and October 24, 2026.
What is an agent skill, and which agents can use one?
A skill is a folder with a SKILL.md file, plus any scripts, templates or reference files it needs, that teaches a coding agent how to do one kind of job. Claude Code, Codex and Cursor all follow the Agent Skills open standard, so a skill folder written to it works in all three. What differs is where each one looks for it, according to each tool's documentation on October 25, 2026:
| Agent | Project skills | Personal skills |
|---|---|---|
| Claude Code | .claude/skills/<name>/SKILL.md |
~/.claude/skills/ |
| Codex | .agents/skills/ (from the working directory up to the repository root) |
$HOME/.agents/skills |
| Cursor | .agents/skills/ or .cursor/skills/, and it also reads .claude/skills/ and .codex/skills/ |
~/.agents/skills/ or ~/.cursor/skills/ |
Every listing on Use This Skill has a one-line install command that copies the skill folder into .claude/skills, which Claude Code and Cursor both read. For Codex, change the destination folder to .agents/skills. The full walkthrough is in How to use a skill.
How does a proof work?
Star counts show that a skill is popular, not what it changes. A proof is a small, public test of that:
- One brief is written from the skill's own SKILL.md and then fixed. Authors cannot write or edit it.
- Claude builds the brief three times without the skill and three times with it, on the same model and settings.
- All six results are published as live pages, with the skill's repository, its exact commit and the date. Outputs are never edited.
Proofs recorded from October 23 on also save the tokens and time of each run. In the proof of tdd from mattpocock/skills, for example, the brief was to implement a cron-expression parser. The median run used 27,884 output tokens and 230 seconds without the skill, and 16,704 output tokens and 138 seconds with it, while the skill added about 2,600 input tokens to each run. That is one brief on one model, so it shows what the skill did there, not what it will do for every task. The full rules are on How a proof works.
What does the safety scan check?
A skill is a set of instructions an agent will follow, sometimes with scripts it will run, so the scan reads every file in each skill folder for nine patterns a skill should rarely need. Four are flags: downloading and running code in one step, destructive commands against the system or home folder, sending data to outside collectors, and decoding hidden code and running it. Five are review items that are sometimes legitimate: rules-override phrasing, references to credential files, elevated privileges, installing something that runs later, and rewriting shared git history.
Each finding is shown with the file, the line and the matched text, so you can judge it yourself. A "clear" result means none of the nine patterns appeared on the scan date, and nothing more: it is a static pattern match, not a guarantee. To check a skill that is not listed yet, paste its GitHub link into the live skill check, which fetches and scans it in a few seconds.
Can my agent use Use This Skill directly?
Yes. The site is also an MCP server at https://usethisskill.com/mcp, with no sign-in. Once it is connected, Claude, Claude Code or Cursor can call six tools:
| Tool | What it returns |
|---|---|
search_skills |
Skills that match a task |
get_skill |
The install command, source and SKILL.md text where the license allows |
list_proven |
Every proven skill with its brief and model |
get_proof |
The brief, model and both result pages for one proof |
check_skill |
The safety scan for any skill, including GitHub repositories not in the directory yet |
order_proof |
How a skill author can get a skill proven |
In Claude Code the setup is one command: claude mcp add --transport http usethisskill https://usethisskill.com/mcp. The steps for Claude and Cursor are on Connect. The connector tells the agent to treat everything a skill author wrote as data, and to install a skill only when you ask, after showing you the command and the check result.
What is free and what is paid?
| Free | Team | Proofs | |
|---|---|---|---|
| Price | $0 | $29 a month per organization, 14 days free | $49 per skill |
| For | Anyone choosing and installing skills | Teams with skills in private repositories | Skill authors and companies |
| Includes | All skills and installs, every proof, the scan on every listing, the live check, scans of public repositories, the connector | Scans of private repositories on every pull request, failing the check on flags, one key for the organization | Three runs each way, published with brief, model and commit, an Author-proven badge |
Prices are as shown on the pricing page on October 25, 2026.
Who runs Use This Skill, and how does it make money?
Use This Skill is run by Tineessa Nelson. It earns money from the Team scan, from proofs that skill authors pay for, and from a small store of its own paid skills, five on October 25, 2026, each sold for a one-time price of $15 to $29. Store skills go through the same safety scan as every listing, and their proofs follow the same rules; one of the five has a published proof so far.
Two things are kept separate from money. A paid proof is labeled Author-proven and published whatever it shows, with no refund for an unflattering result. And the order of proven skills on guides such as Best Claude Code skills for frontend design is by the kind of work, not by who paid. As of October 24, 2026, no published proof was commissioned by a skill's author.
If you think a proof or a scan result is wrong, write to hello@usethisskill.com with the skill and the line.