Prompts to Build Your Own Claude Skill
A Claude skill teaches Claude to do one task your way, every time, without you pasting the same instructions again. But a skill only works if it's built right. With a vague description, Claude rarely uses it. With bloated instructions, it can miss steps. These 10 prompts help you decide if a task needs a skill, write one, and test and fix it.
Each one started as a draft from Promptimize, then was edited by hand. Every prompt follows Anthropic's published rules for skills, uses only the details you give it, and is told not to make up facts.
Want to see how they hold up? Paste any of them into our free Prompt Grader for a score out of 100 and a breakdown of what's working.
They work in ChatGPT, Claude, and Gemini.
How to use these prompts
Copy a prompt into Claude, ChatGPT, or Gemini and paste your details where marked. When you have a finished SKILL.md, put it in a folder with the same name as the skill, zip the folder, and upload it in Claude under Settings, then Capabilities (this needs a paid plan with code execution turned on).
Plan and build your skill
1. Decide if a task should be a skill
Use it when: You repeat a task with Claude but aren't sure a skill is worth it.
Why it works: It asks about your task first and is told to recommend a plain saved prompt when that would work just as well.
Act as a workflow consultant who helps non-technical people set up repeat work with Claude. I do the task below over and over, and I can't tell whether it should become a Claude skill or stay a saved prompt I paste in. For context: a skill is a folder with a SKILL.md file. Claude sees only its name and description until a request matches, then loads the full instructions. Skills fit best when a task comes up often, follows the same steps each time, and needs knowledge or rules I would otherwise re-explain. 1. Ask me up to 5 short questions about how often I do this, whether the steps change, and what I always have to explain. Wait for my answers. 2. Weigh the answers before deciding. 3. Give a one-line verdict: Skill or Saved prompt. 4. Explain why in 3 to 5 bullets, then give my next step in one sentence. Rules: Don't recommend a skill just because it sounds more advanced. If a saved prompt works just as well, say so. The task I repeat: [describe the task in a sentence or two]
2. Get interviewed, then get a complete SKILL.md
Use it when: You know the task but don't know how to write a skill.
Why it works: It asks before it writes, checks its summary with you, and builds the file to Anthropic's name and description rules.
Act as an experienced skill author who turns everyday tasks into Claude skills for people with no coding background. Interview me about the task below, then write a complete SKILL.md file. 1. Ask me 1 or 2 questions at a time about the goal, the exact steps, the result I expect, and what usually goes wrong. Keep going until you understand the task. 2. Summarize what you heard in 5 bullets and wait for me to confirm. 3. Write the file. It must start with frontmatter containing: - name: lowercase letters, numbers and hyphens only, 64 characters max, without the words claude or anthropic. Prefer a form like writing-weekly-updates. - description: third person, under 1024 characters, no XML tags, saying what the skill does and when to use it. 4. Then a short body: a one-line purpose, numbered steps, and a checklist Claude runs before finishing. 5. Check the file against these rules and fix anything that breaks them. Rules: Use only details I give you. Where something is missing, write (FILL IN: what is needed) instead of guessing. Keep the body short and include only what Claude wouldn't already know. The task I want a skill for: [describe the task]
3. Turn a prompt you keep reusing into a skill
Use it when: You paste the same long prompt into chat again and again.
Why it works: It keeps what is specific to your task, cuts filler, and lists every change so you can see nothing important was lost.
Act as a prompt engineer who turns reusable prompts into Claude skills. I keep pasting the same long prompt into chat. Turn it into a skill so Claude follows it automatically. 1. Read my prompt and pick out its goal, steps, rules, and output format. 2. Ask me when I usually use it, so the description matches those moments. Wait for my answer. 3. Keep everything specific to my task. Cut lines that only tell the AI to be generally helpful or careful. 4. Write a complete SKILL.md: frontmatter with a name (lowercase letters, numbers and hyphens, 64 characters max, no claude or anthropic) and a third-person description under 1024 characters that says what it does and when to use it. Then the instructions as numbered steps. 5. Compare the skill to my original prompt and confirm nothing important was lost. Rules: Don't add requirements or facts that aren't in my prompt. After the file, list what you removed or reorganized and why, one line each. The prompt I keep reusing: [paste your prompt]
4. Turn your checklist or process into a skill
Use it when: You have a written process, SOP, or checklist you want Claude to follow exactly.
Why it works: It keeps your steps in order, asks about anything vague, and adds a check-and-fix step where mistakes are costly.
Act as a process documentation expert who turns step-by-step procedures into Claude skills. I want Claude to follow my process the same way every time. 1. Read my process and list its steps, decision points, and required outputs. 2. Ask me about any step that's vague or depends on knowledge only I have. Wait for my answers. 3. Mark the steps where a mistake is costly, and add a check: Claude verifies the step, fixes any problem, and checks again before moving on. 4. Write a complete SKILL.md: frontmatter with a name (lowercase letters, numbers and hyphens, 64 characters max, no claude or anthropic) and a third-person description under 1024 characters that says what it does and when to use it, then my steps as a numbered workflow with a closing checklist. 5. Confirm every original step is included and nothing was added beyond the checks from step 3. Rules: Keep my steps in their original order. Apart from those checks, don't invent steps, tools, or rules. If something is unclear, ask instead of guessing. My process (a checklist, SOP, or notes are fine): [paste your process]
Make it trigger and work well
5. Write a description so Claude uses your skill
Use it when: You're writing or rewriting the description line of a skill.
Why it works: It builds the description from the words real people type, and checks it against requests that should not trigger it.
Act as a specialist in writing Claude skill descriptions. Claude only sees a skill's name and description until it decides to use it, so the description decides whether my skill gets used at the right time. 1. Summarize what my skill does in one plain sentence. 2. List 5 to 8 realistic phrases or situations where someone would need it, in the words a real person would type. 3. List 2 or 3 similar requests that should not trigger it. 4. Write 2 candidate descriptions using the words from step 2. Each must be in third person (for example, "Drafts weekly updates..."), say what the skill does and when to use it, stay under 1024 characters, and contain no XML tags. 5. Show each with its character count, and recommend one in a sentence. Rules: Don't claim the skill does anything I haven't described. Avoid vague wording that could match unrelated requests. What my skill does and when I use it: [paste here, or paste your SKILL.md]
6. Fix a skill that triggers at the wrong times
Use it when: Your skill isn't used when it should be, or kicks in when it shouldn't.
Why it works: It diagnoses each misfire from your real examples before rewriting, then gives you test requests to confirm the fix.
Act as a troubleshooter for Claude skills. My skill either isn't used when it should be, or is used when it shouldn't be. Claude decides from the name and description alone, so that's where the fix usually belongs. 1. For each example below, explain in one sentence why the description led Claude to the wrong decision. 2. List the trigger words or situations that are missing, and any wording that's too broad. 3. Rewrite the description in third person, under 1024 characters, with no XML tags, saying what the skill does and when to use it. 4. Show it in a code block with its character count. 5. Give me 5 test requests: 3 that should use the skill and 2 that shouldn't. Rules: Base the fix only on my examples and what the skill actually does. Don't claim abilities it doesn't have. My skill's name and description: [paste here] Requests where it misfired (and whether it should have been used): [paste 2 or 3 examples]
7. Add an example without made-up details
Use it when: Your skill works, but its output style isn't consistent.
Why it works: It builds the example only from your real sample and uses (FILL IN) for values that change, so Claude never copies a made-up number.
Act as a skill editor who improves Claude skills with accurate examples. My skill works, but the style of its output isn't consistent. I want to add an example or template built only from real material I provide. 1. Find the part of my skill where style or format matters most. 2. Study the tone, structure, and formatting of my real sample. 3. Choose between a worked example (my sample, lightly trimmed) and a fill-in template, and explain the choice in one sentence. 4. Write the new section using only content from my sample. In a template, or anywhere a detail would change each time, like a name, date, price, or number, write (FILL IN: what goes here) instead of a specific value, so Claude never copies it as a default. 5. Keep it short, and tell me exactly where to place it in my SKILL.md. Rules: Every detail must trace back to my sample. Don't invent names, numbers, prices, or facts. My SKILL.md: [paste your file] A real example of the output I want: [paste a real sample]
Check and maintain your skill
8. Review your skill against Anthropic's rules
Use it when: You've written a SKILL.md and want it checked before you upload it.
Why it works: It checks each rule one at a time, separates hard rule breaks from suggestions, and hands back a corrected file.
Act as a careful reviewer of Claude skills who knows Anthropic's skill authoring rules. Check my SKILL.md and fix it. Check these in order and mark each Pass or Fail: 0. Frontmatter: the file starts with a frontmatter block that has both a name and a description. 1. Name: 64 characters max, only lowercase letters, numbers and hyphens, and no claude or anthropic. 2. Description: under 1024 characters, third person, no XML tags, and says both what the skill does and when to use it. 3. Length: the instructions are under 500 lines. 4. Content: only task-specific instructions, one default approach instead of a menu of options, and a checklist or steps where they help. 5. Examples: none contain made-up numbers, names, or facts that Claude might copy as defaults. 6. Reference files, if any: linked directly from SKILL.md, not from other reference files. Give a table with four columns: Rule, Hard rule or suggestion, Pass/Fail, Fix. Checks 0 to 3 are hard rules; the rest are suggestions. Then the corrected SKILL.md in one code block, and a one-line summary of the changes. Rules: Don't change what my instructions mean, and don't add facts I didn't give you. Where a fix needs information you don't have, write (FILL IN: what is needed). My SKILL.md: [paste your file]
9. Test your skill before you rely on it
Use it when: You've uploaded a skill and want to know it actually works.
Why it works: It writes realistic requests, including ones that should not trigger the skill, and compares results with the skill on and off.
Act as a tester for Claude skills who writes simple test plans anyone can run in a normal chat. I want to check that my skill works before I rely on it. 1. Summarize what my skill does and when it should be used. 2. Write at least 3 realistic requests that should use it, phrased the way a real person would type them, plus 1 tricky edge case and 1 request that should not use it. 3. For each, write 2 or 3 things a good result must do, based only on my skill's instructions. 4. Give me a table: Request, Should use the skill (yes/no), What to look for. 5. Below it, a 3-step plan: try each request with the skill on, try the same requests with it off, then compare and note what to fix. Rules: Don't test for rules my skill doesn't contain. Keep everything doable without code. My SKILL.md: [paste your file]
10. Split a skill that has gotten too long
Use it when: Your SKILL.md keeps growing and Claude starts missing steps.
Why it works: It keeps the steps Claude needs every time in the main file and moves the rest into clearly named files Claude reads only when needed.
Act as a skill architect who reorganizes overgrown Claude skills. My SKILL.md has gotten long. Anthropic recommends keeping it under 500 lines and moving extra detail into separate reference files that SKILL.md links to directly. 1. Outline the sections of my file. 2. Label each one as core (Claude needs it every time) or reference (needed only in some cases, like long examples or edge cases). 3. Group the reference sections into a few clearly named files, like examples.md or edge-cases.md. 4. Rewrite SKILL.md with the frontmatter, the core steps, and a link to each reference file with one line on when to read it. 5. Confirm nothing was lost and no reference file links to another reference file. Show the folder layout first, then each file in its own code block with its filename above it. Rules: Don't change the wording of the content you move, and don't remove anything. The only new text should be the links and their one-line notes. My SKILL.md: [paste your file]
FAQ
What is a Claude skill?
A folder with a SKILL.md file that teaches Claude to do one task your way. Claude reads its name and description, and loads the full instructions only when a request matches.
Do I need a paid Claude plan?
To upload a skill in the Claude app, yes. Skills need a paid plan (Pro or higher) with code execution turned on. You can still run these prompts in any AI to write and review the file.
Should I use these prompts or the Skill Builder?
The Skill Builder writes a complete, ready-to-upload skill in one step. These prompts help with everything around it: deciding if a task needs a skill, fixing one that doesn't trigger, and reviewing or testing one you already have.
More prompt packs
Free, hand-edited ChatGPT prompts to match your resume and cover letter to any job description. Find gaps, rewrite bullets, and prep for the interview.
Free, hand-edited ChatGPT prompts for small business customer emails: angry complaints, late orders, refund requests, apologies, reviews, and more.
Free, hand-edited ChatGPT prompts to prepare for a job interview. Practice your answers, handle the salary question, run a mock interview, and follow up.
Free, hand-edited prompts to write, fix, and test the instructions for your own Custom GPT or Claude Project, plus ready-made assistants to set up.
Free, hand-edited ChatGPT prompts to review a job offer, write a counteroffer, negotiate other terms, and ask for a raise. Uses only your numbers.
Want the whole skill built for you? Describe the task, and the Skill Builder writes a ready-to-upload Claude skill, built to Anthropic's rules. Free, 5 per day.
Try the Skill Builder