Agent Skill Scaffolder
Generate a valid SKILL.md for a Claude Agent Skill — YAML frontmatter (name, description, allowed-tools) plus instructions. Runs in your browser, free.
Agent Skill Scaffolder
How to Use the Agent Skill Scaffolder
Name the skill in kebab-case
Give the skill a short, hyphenated name like pdf-form-filler. The field auto-sanitises to lowercase letters, digits and hyphens — exactly what a skill folder name and the name: frontmatter key require.
Write the description trigger
The description is the single most important line: it is what the model reads to decide whether to load the skill. State plainly when the skill applies — the conditions, keywords, or task shape that should trigger it.
Optionally restrict the tools
List the tools the skill is allowed to call as a comma-separated set — they become an allowed-tools entry in the frontmatter. Leave it blank to inherit the full tool set. This is how you scope a skill to least privilege.
Add instructions, then copy or download
Write the markdown body — the steps the agent should follow once the skill loads. The assembled SKILL.md builds live in the box below; copy it or download the file straight into your skill's folder. Everything runs in your browser.
SKILL.md Is the Contract Between Your Knowledge and the Agent
What an Agent Skill actually is
A Claude Agent Skill is a folder that packages procedural knowledge so an agent can pick it up on demand. At the centre of that folder is a single file — SKILL.md — that combines a small block of structured metadata with a body of plain-English instructions. The metadata sits at the very top of the file inside a YAML frontmatter block, delimited by a line of three dashes (---) before and after it. Inside that block live two required keys: name, the skill's machine identifier, and description, the one-line statement of when the skill should be used. An optional allowed-tools key narrows which tools the skill may invoke. After the closing --- and a blank line comes the Markdown body: the step-by-step guidance the agent reads once the skill is loaded.
The reason this format matters is a design principle called progressive disclosure. An agent cannot afford to hold the full text of every skill it owns in its context window at once — that would be wasteful and would dilute its attention. Instead, the agent first sees only the lightweight metadata: every skill's name and description. From that thin slice it decides which skill, if any, is relevant to the task in front of it. Only then does it pull the full body of the chosen skill into context. This is exactly why the description field carries so much weight: it is the entire basis on which the skill is selected. A vague description — "helps with documents" — leaves the agent guessing; a precise one — "use when the user needs to fill or flatten a fillable PDF form from supplied field values" — tells it precisely when to reach for the skill and when to leave it alone.
"The body of a skill is only ever read after the description has earned its place. Write the description as if it is the only line the agent will see — because, most of the time, it is."
Why a generator beats hand-typing the frontmatter
The frontmatter looks trivial, but it is YAML, and YAML has sharp edges. A description that happens to contain a colon — say, "use when: the file is a PDF" — will break the parser unless the value is wrapped in quotes, because the colon reads as a key/value separator. The same applies to values starting with a hyphen, containing a hash, or holding other reserved characters. This tool handles that for you: it inspects each value and double-quotes it only when a special character demands it, escaping any embedded quotes along the way, so the file always parses. It also enforces kebab-case on the name as you type, omits the allowed-tools line entirely when you leave the field blank (rather than emitting an empty key), and guarantees the blank line between the closing --- and the body that the format expects. The result is a SKILL.md that is valid the moment you copy it. Drop the file into a folder named after your skill, place that folder where your agent looks for skills, and the agent can discover it, read its description, and — when the task matches — load its instructions. Because the whole thing assembles in your browser with plain JavaScript, nothing you type about your internal tooling, prompts, or workflows is ever sent to a server or a model. You get a clean, spec-shaped starting point, and you keep full control over what you eventually ship.
10 Facts About Agent Skills
A skill is just a folder; its heart is a single SKILL.md file.
The YAML frontmatter at the top is delimited by a line of three dashes (---) above and below it.
Only two keys are required: name and description.
The description is what the agent reads to decide whether to load the skill at all.
Progressive disclosure means metadata loads first; the full body only loads once the skill is chosen.
An optional allowed-tools key scopes a skill to least privilege.
The name is kebab-case — lowercase letters, digits and hyphens — matching its folder.
A description containing a colon must be quoted, or the YAML parser misreads it.
The body is plain Markdown — the steps the agent follows once the skill is in context.
This scaffolder runs entirely in your browser — your skill details are never uploaded.
Frequently Asked Questions
- It is the central file of a Claude Agent Skill. It begins with a YAML frontmatter block — delimited by lines of three dashes — that holds the skill's name and description (and optionally an allowed-tools list), followed by a Markdown body of instructions the agent reads when the skill is loaded.
- Because of progressive disclosure: the agent first sees only each skill's name and description, and uses the description alone to decide whether to load the full skill. A precise description that states exactly when the skill applies is the single biggest factor in whether the skill gets used correctly.
- A skill's name is also its folder name and machine identifier, so it must be lowercase letters, digits and hyphens only. The tool sanitises anything you type — spaces, capitals and punctuation — into that shape automatically as you go.
- No — it is optional. Leave the field blank and the skill inherits the full available tool set, and the line is omitted from the frontmatter entirely. Fill it with a comma-separated list to scope the skill to just those tools (least privilege).
- It is a small block of structured metadata at the very top of the file, written in YAML and fenced by a line of three dashes (---) before and after. In a SKILL.md it carries the name, description and optional allowed-tools keys, separated from the Markdown body by the closing dashes and a blank line.
- A bare colon would break the YAML parser, which would read it as a key/value separator. The tool detects colons and other special characters and wraps the value in double quotes (escaping any quotes inside), so the file always parses cleanly.
- No. The SKILL.md is assembled entirely in your browser with plain JavaScript. Nothing you type — the name, description, tools or instructions — is sent to any model, server or third party, and nothing is stored.
- Save it as SKILL.md inside a folder named after your skill (matching the name field), then place that folder wherever your agent discovers skills. The agent reads the frontmatter to find it and loads the body when the task matches the description.
- The procedural knowledge the agent should follow once the skill loads: the steps to take, the order to take them in, any conventions or edge cases to watch for. Plain Markdown works well — headings and numbered steps keep it readable for both the agent and you.
- No — it is an independent helper that produces a correctly shaped SKILL.md. Always check the current Agent Skills documentation for any keys or conventions specific to your setup; this tool gives you a valid, well-formed starting point you can extend.
- Completely free, with no account or sign-up and no usage limit. It runs in your browser and collects no data.
Related News
You may be interested in these recent stories from our newsroom.
No related news yet for this tool. Our editorial team publishes new pieces every week.
Browse all news →75 more free tools
Calculators, converters, security tools — no signup.