Free tool
SKILL.md format checker: read any GitHub repository
A skill is a directory with one Markdown file in it, and the file opens with YAML frontmatter. When that frontmatter is wrong the failure is quiet: the skill loads with empty metadata and never triggers, or it works locally and is refused the first time the folder is uploaded or packaged.
Paste a repository URL, or a direct link to a skill folder, and this page reads every SKILL.md it finds and reports what the specification says about each one. It is a format check and nothing more.
- Publisher
- Skills Board
- Price
- Free, no account
- Checked against the spec
Check a repository
Loading the checker.
What it checks
Two levels, kept apart on purpose. An error is a rule the Agent Skills specification states. A warning is a convention Anthropic's authoring guidance recommends and the specification does not require, so a file with warnings and no errors is still a valid file.
- The file parses at all
- That the file opens with a --- frontmatter marker, that the block is closed, that the YAML inside it parses, and that no key is declared twice. A file that fails here loads with empty metadata, which is the failure nobody notices.
- The two required fields
- That name and description are present and are strings. An unquoted value such as true, no, or 1.0 parses as a boolean or a number rather than as the text that was typed.
- The limits the specification states
- The name is 1 to 64 characters of lowercase letters, numbers, and hyphens, with no leading, trailing, or doubled hyphen. The description caps at 1024 characters, compatibility at 500.
- The name against the directory
- The specification requires the name field and the folder holding SKILL.md to carry the same string. A mismatch passes where an agent reads the folder from disk and fails on upload or packaging.
- Fields the format does not define
- A key outside the six the specification defines is reported as a warning: Claude Code ignores it, and a claude.ai upload or a packaging script refuses the folder with an unexpected-key error.
- The shape of the rest
- That metadata is a map of strings, that allowed-tools is the space-separated string the specification defines, and that the body is not empty and stays inside the recommended length.
What it does not check
The check reads format. Everything below is outside it, and reading a green report as any of these would be reading it wrong.
- It is not a review
- Nothing here judges whether the instructions in the body are correct, useful, or well written. A file can pass every rule and still be a bad skill.
- It is not a security audit
- The body is not inspected for what it asks an agent to do, and no script, reference, or asset beside SKILL.md is read or run. Decide for yourself whether to trust a skill before you install it.
- There is no score
- The report is a list of specific findings with the rule behind each one. There is no rating, no grade, and no ranking of one repository against another.
- The repository stays its authors'
- This page reads public files and quotes what it found. The skills belong to the people who published them, under whatever license they chose.
Reading the report from an agent
The page and the endpoint take the same parameter, so a report is a link. Addformat=mdfor the same report as Markdown, or send an Accept header asking for text/markdown.
https://www.skillsboard.sh/api/check?url=https://github.com/owner/repo&format=mdKeep going
- Write a valid SKILL.md in the browser
The same rules from the other side: fill in the fields, watch the checks, download the folder. A report opens a checked file straight into it.
- How to write a SKILL.md
The guide behind these checks: every field, what it does, and the wording that makes a skill trigger.
- Agent Skills: the open standard
What the specification defines, which agents implement it, and where each one looks on disk.