Check your SKILL.md
Free. Checks a skill against the Agent Skills spec and Anthropic's best practices, and shows the two lines an agent reads before it loads it.
Valid, but it may not trigger
2 warnings, 1 note
- name
- 14 / 64
- description
- 16 / 1,024
- body lines
- 4 / 500
Drop a file here, or paste it. The rule checks run in your browser and nothing is stored.
pdf-processing
Helps with PDFs.
At the start of a session an agent sees only these two fields for each installed skill, and opens the rest of the file when a request matches.
Warning: description does not say when to use the skill
Agents load a skill by matching the request against its description. Add a trigger such as "Use when the user asks to…" with the words people actually type. Best practices
Warning: description is very short
Short descriptions such as "Helps with PDFs" rarely trigger reliably. Name the tasks, file types and keywords the skill covers. Spec
Note: Folder name must match
Save this file as pdf-processing/SKILL.md: the spec requires name to match the folder that holds it. Spec
AI review
Scores how reliably the description makes an agent load the skill and how usable the instructions are, with a fix for anything weak.
What it checks
Spec rules, on the left. Break one and the skill will not load.
Best practices, on the right. They decide whether the skill triggers.
Frontmatter
Opens and closes with a line of three hyphens. The YAML between them must parse.
A legal name
Up to 64 lowercase letters, numbers and hyphens, the same as its folder. No "claude" or "anthropic".
A description that fits
1 to 1,024 characters with no XML tags. This one is 131.
---name: invoice-tablesdescription: Extracts tables, line items and totals from PDF invoices into CSV. Use when the user asks to pull data out of an invoice or a bill.--- # Invoice tables Run scripts/extract.py on the PDF, then check that the line items add up to the total. For column names and formats, see [reference/columns.md](reference/columns.md).
What it does, and when
In the third person, naming the tasks and the words people type: "Use when the user asks to…"
A short body
Under 500 lines and about 5,000 tokens. The agent reads it only once the skill loads.
Portable paths
Longer material lives in files linked by relative paths with forward slashes.
How a check runs
Rules first, in your browser. Then an optional AI review. Then the only test that counts: your agent loading the skill.
First, paste your SKILL.md
The spec and best-practice rules run on every keystroke. Errors and warnings appear beside the editor, each with the rule it comes from.
You paste
--- name: pdf-processing description: Helps with PDFs. --- # PDF processing Use pdfplumber to extract text from the PDF.You get back, instantly
- Folder name must match
- description does not say when to use the skill
- description is very short
Then run the AI review
It scores how reliably the description makes an agent load the skill, and suggests a fix for each weak score. Add a request to test it against.
1 review is free without an account; a free account includes 3 a month.
You type a request
pull the tables out of this invoice PDF
You get back
29/100Likely skipped for this requestExample result
- Tells the agent when to load it22
- Says what it does and when30
- Specific about tasks and keywords18
- Instructions are actionable64
- Instructions are concise88
- Instructions match the description57
Last, try it in your agent
Every agent reads the same folder from its own place. Start a new session and ask for the task in your own words.
You run
mkdir -p ~/.claude/skills/pdf-processing cp SKILL.md ~/.claude/skills/pdf-processing/Then you ask
“pull the tables out of this invoice PDF”Claude Code opens pdf-processing when the request matches its description. If it does not, the description is what to fix.
Rather not write one?
Browse ready-made skills for Claude Code, Codex, Cursor and Gemini CLI, or read what Claude skills are first. Publishing your own? See how to publish on SkillGild.
SKILL.md checker questions
What it checks, why skills fail to trigger, and what happens to what you paste.
Still have a question?Search help topics or send the team a request.It validates the YAML frontmatter against the Agent Skills specification: name is 1 to 64 lowercase letters, numbers and single hyphens without reserved words, description is 1 to 1,024 characters, and optional fields such as compatibility and metadata have the right shape. It also flags best-practice problems from Anthropic's authoring guide, such as a description that never says when to use the skill, first-person wording, or a body over 500 lines.