Skills
Folder with SKILL.md plus scripts and references. Loaded only when a request matches its description.
Free Agent Skills tool
Free Claude Skill Generator
A Claude skill, also called an Agent Skill, is a folder with a SKILL.md file that teaches Claude a repeatable task: YAML frontmatter with a name and a trigger-focused description, short instructions, and optional scripts, references, and assets that Claude loads only when needed. This generator turns a plain-English description of your task into a valid SKILL.md, lints it against the documented limits, and packages it as a .zip for Claude Code, Claude.ai, or the Claude API.
Runs in your browserNo signupLinted against the docsSKILL.md and .zip
Starter templates
Start from an example skill
Output
reviewing-code-changes/SKILL.md
0 errors0 warnings2 info
--- name: reviewing-code-changes description: Reviews code changes against the team checklist and reports bugs, security risks, and missing tests ranked by severity. Use when the user asks for a code review, a PR review, feedback on a diff, or a check of staged changes before merging. allowed-tools: Read Grep Glob Bash(git diff *) --- # Code Review Checklist Reviews code changes against the team checklist and reports bugs, security risks, and missing tests ranked by severity ## When not to use - The user wants new code written from scratch - The change is documentation only ## Steps 1. Collect the change: run `git diff --staged`, or `git diff main...HEAD` when nothing is staged. 2. Read each changed file in full where the diff alone hides context. 3. Check every item in references/checklist.md and note the file and line for each finding. 4. Rank findings as Blocker, Should fix, or Nit, and drop anything you cannot point to in the code. 5. Report findings first, then a one-line verdict: approve, approve with nits, or request changes. ## Guidelines - Cite file:line for every finding - Suggest a concrete fix, not just the problem - Do not comment on formatting a linter already enforces ## Example **Input** ```text Review my staged changes before I open the PR. ``` **Output** ```text Blocker: src/api/orders.ts:42 trusts req.body.userId; use the session user instead. Should fix: no test covers the empty cart path. Verdict: request changes. ``` ## Supporting files - [references/checklist.md](references/checklist.md): The full review checklist. Check every item.
Body: 42 of 500 recommended lines.
Folder layout
reviewing-code-changes/
├── SKILL.md
└── references/
└── checklist.mdSkill linter
- Info: Description is 239 characters. The Claude.ai help center lists a 200-character maximum for uploads; the API and spec allow 1024.
- Info: allowed-tools is a Claude Code field. Claude.ai and the API ignore it.
Install the skill
1. Personal skill (all projects)
Unzip into ~/.claude/skills so the file lands at ~/.claude/skills/reviewing-code-changes/SKILL.md.
mkdir -p ~/.claude/skills unzip reviewing-code-changes.zip -d ~/.claude/skills
2. Project skill (commit it with the repo)
Run from the repository root so the file lands at .claude/skills/reviewing-code-changes/SKILL.md.
mkdir -p .claude/skills unzip reviewing-code-changes.zip -d .claude/skills
3. Use it
Claude loads the skill when a request matches its description. Run /reviewing-code-changes to invoke it directly.
How Claude skills work
Skills use progressive disclosure, so you can install many of them without filling the context window.
1. Metadata, always loaded
The name and description from every skill sit in the system prompt, about 100 tokens each. Claude matches requests against the description.
2. SKILL.md, loaded on trigger
When a request matches, Claude reads the SKILL.md body. Keep it concise and under 500 lines.
3. Files, loaded as needed
Claude reads reference files and runs scripts only when the steps call for them. Script code never enters context, only its output.
SKILL.md format reference
| Field | Applies to | Rule |
|---|---|---|
| name | All surfaces (required by the API and spec) | Max 64 characters. Lowercase letters, numbers, and hyphens. No XML tags. Cannot contain "anthropic" or "claude". |
| description | All surfaces (required) | Non-empty, max 1,024 characters, no XML tags. Third person. Says what it does and when to use it. |
| license, compatibility | Agent Skills spec | Optional. compatibility is max 500 characters. |
| allowed-tools | Claude Code | Tools pre-approved while the skill is active, such as Read Grep Bash(git diff *). |
| argument-hint, model | Claude Code | Autocomplete hint like [issue-number], and a model override or inherit. |
| disable-model-invocation, user-invocable | Claude Code | Control whether Claude can load the skill automatically and whether it shows in the / menu. |
| SKILL.md body | All surfaces | Keep it under 500 lines. Link reference files one level deep with relative forward-slash paths. |
Sources, checked October 28, 2026: Agent Skills overview, Skill authoring best practices, Claude Code skills, Agent Skills specification.
Skills vs CLAUDE.md vs subagents vs slash commands
CLAUDE.md
Project memory loaded into every session. Best for conventions that always apply.
CLAUDE.md GeneratorSubagents
Separate assistants with their own context window and tools for delegated work.
Subagent GeneratorSlash commands
Prompts you run on demand with /name. Skills can be invoked the same way and also trigger on their own.
Slash Command GeneratorHow to create a Claude skill
- 1Pick a template or start freshLoad a starter such as code review, release notes, or CSV cleanup, or write your own skill from scratch.
- 2Name it and write the triggerEnter a title, keep the generated lowercase name, and describe what the skill does and when Claude should use it.
- 3List the steps and add filesWrite the ordered steps, guidelines, and an example, then add reference files, scripts, or assets.
- 4Fix lint issues and downloadResolve errors in the linter, then copy SKILL.md or download the .zip and follow the install steps for your target.
Claude skills FAQ
What are Claude skills?
Claude skills (Agent Skills) are folders with a SKILL.md file of instructions plus optional scripts, references, and assets. Claude sees only each skill's name and description until a request matches, then loads the instructions and reads the extra files only when needed.
How do I install a skill in Claude Code?
Unzip the skill so its SKILL.md lands at ~/.claude/skills/<name>/SKILL.md for every project, or at .claude/skills/<name>/SKILL.md inside a repository to share it with the team. Claude uses it when a request matches the description, and you can also run /<name>.
What is the SKILL.md description limit?
The description must be non-empty, at most 1,024 characters, and free of XML tags. Write it in third person and say both what the skill does and when to use it, because Claude picks skills by matching requests against this text. The Claude.ai help center lists a shorter 200-character limit for uploads.
Can I use the same skill in Claude.ai and the API?
Yes, the same folder works on each surface, but skills do not sync. Upload the zip in Claude.ai under Customize > Skills, upload the files to the /v1/skills API for your workspace, and copy the folder into a skills directory for Claude Code. Claude Code-only fields such as allowed-tools are ignored elsewhere.
Why can't my skill name contain "claude"?
Anthropic reserves the words "anthropic" and "claude" in skill names, so a name such as claude-helper is rejected. Names must also be 64 characters or fewer and use only lowercase letters, numbers, and hyphens. Name the activity instead, for example reviewing-code-changes.
Keep building
Run your skills inside deterministic workflows
Tornic chains Claude Code and Codex steps into repeatable multi-agent workflows, so a skill like code review or release notes runs the same way every time.
Explore TornicTornic is not affiliated with or endorsed by Anthropic. Claude is a trademark of Anthropic, PBC. Check generated skills before using them, and only install skills from sources you trust.