← Back to all free guides
A free guide by Mariah Brunner ↗ INSTAGRAM ↗ TIKTOK

5 Skill Hacks
Nobody Talks About

You’ve built a few skills. They kind of work. These are the tricks that make them work every time.

Before you build anything from scratch, check what already exists. Most people skip this and rebuild what someone already made.

  1. Anthropic’s skills. Go to Customize (renamed Plugins on newer accounts), then Skills, and add pre-made ones from Discover. Skills work on every plan, including Free. Turn on Create and edit files (older accounts call it Code execution and file creation) under Settings, then Capabilities, first.
  2. Plugins come with skills. A plugin bundles skills and connectors for one job, like sales or marketing. On paid plans, open the Plugins tab, then Discover, pick one and click Add. (How plugins work)
  3. My 10 work skills. I built 10 skills for your 9 to 5 you can copy right now, including a weekly status report, a meeting recap and an email draft.

New to skills? Start with Build Claude Skills That Actually Work. This guide is about getting more out of the ones you have.

The 5 Hacks

01

Debug the trigger in 2 minutes

Skill not running when it should? Don’t guess. Claude picks a skill by matching your request against its description, so that’s almost always where the problem is. Paste this and let Claude show you what it sees.

Debug My Skill
I have a skill called [SKILL NAME] that isn't running when I expect it to. Help me fix it. 1. Read me the skill's name and description exactly as you see them. Don't paraphrase. 2. Here are 3 things I typed where it should have run: [PASTE 3 REAL REQUESTS]. For each one, tell me honestly whether the description would make you pick this skill, and why or why not. 3. Here's 1 thing where it ran but shouldn't have (or say "none"): [PASTE]. 4. Rewrite the description so it fires on all 3 and skips the wrong one. It must say what the skill does AND when to use it, include the words I actually type, and stay under 200 characters. Show me the old and new description side by side, with the character count. Don't change the skill's instructions, only the description.

To check if a skill ran, look in Claude’s thinking for "Reading" plus the skill’s name. (Anthropic’s skills tutorial)

02

Add a do-not-use-for line

A lot of broken skills aren’t broken. They’re firing at the wrong time. Without a boundary, your email skill jumps in every time you mention email. Build your description like this and keep it under 200 characters.

Description Template
[What the skill does, in one line]. Use when I ask for [the phrases I actually say]. Do not use for casual chat, brainstorming or anything unrelated to [the specific task].

03

Turn any great chat into a skill

You don’t have to build skills from scratch. When a chat goes really well, paste this at the end. Claude reads back over the whole conversation and packages it. Then upload the ZIP under Customize or Plugins, then Skills, +, Create skill, Upload a skill.

Chat To Skill
That workflow worked. Turn what we just did into a reusable skill. 1. Look back over this whole chat. List the steps that worked, the format I approved, and anything I rejected or corrected, so the skill avoids it. 2. Ask me up to 3 questions about anything you'd otherwise have to guess (inputs, audience, edge cases). 3. Write the skill: a lowercase-hyphen name, a description under 200 characters that says what it does and when to use it (with the words I'd actually type), then step-by-step instructions, the output format, and one example from this chat. 4. Package it as a ZIP holding one folder named after the skill with a SKILL.md inside, so I can upload it. 5. Test it on a made-up example and show me the result before you hand it over.

04

Install what you’ll use, without worrying

People hold back because they think extra skills slow Claude down. Until a skill is used, only its name and description load, about 100 tokens each, per Anthropic’s docs. The full instructions load only when your request matches.

So install anything you’d actually use. Two rules. Only install skills from sources you trust, because a skill can tell Claude to run code. And don’t install three skills that do the same job, or Claude has to guess between them.

05

Fix lazy output in your message

If Claude starts rushing through steps inside a skill, don’t rewrite the skill first. Ask for the full version in your message, and make it say what you’re leaving out instead of cutting quietly.

Full Effort Nudge
For this one, take your time and give me the complete version. Don't skip or merge any steps from the skill, and don't summarize where I asked for detail. If something would take too long, tell me what you're leaving out instead of quietly cutting it. When you're done, list each step of the skill and confirm you did it.

The skill sets what to do. Your message sets how much. If you need that nudge every time, then add it to the skill.

Best Practices Most People Miss
  1. Test it right after you build it. Open a new chat, ask for the task the way you normally would and check the output against an example you liked.
  2. Call it by name when it matters. Saying "use my status-report skill" is the most reliable way to make it run.
  3. Be specific. "Write a 3-paragraph email in a warm, professional tone" beats "write an email."
  4. Update your top skills monthly. Your workflows change, so a skill you wrote in spring can be stale by fall.
  5. Use skills inside Projects. A skill running in a Project also gets that Project’s files and instructions, so the output fits your work better.
  6. Chain them. Run your inbox triage skill first, then your meeting prep skill on anything flagged urgent.

Still stuck? Read Why Your Claude Skill Isn’t Working for the most common reasons and how to fix each one.