What is an AI Skill?
An AI skill is a reusable set of instructions that teaches an AI how to do a specific job in a specific way. A prompt asks once. A skill installs a repeatable behaviour.
Prompt vs Skill
The easiest way to understand skills is to compare them with normal prompts. Both are useful. They just solve different problems.
Prompt
A prompt is a single instruction for a single moment.
Write me a LinkedIn post about AI governance.
- Useful for quick tasks.
- Can vary wildly depending on wording.
- Usually does not include testing, guardrails or reusable structure.
Skill
A skill is a reusable operating pattern for a recurring job.
When I say "Governance Post", create a LinkedIn post using: - audience: senior technology leaders - tone: practical, sharp, not fluffy - structure: hook, risk, example, action - guardrail: no fake stats - quality check: flag weak claims before final output
- Better for repeatable work.
- Includes rules, examples, outputs and checks.
- Can be installed into Claude Projects, ChatGPT Projects, Custom GPTs, Gemini Gems or local workflows.
What goes inside a good skill?
A strong skill is not just a mega-prompt wearing a fake moustache. It needs enough structure for the AI to behave consistently and enough checks to catch rubbish output.
Trigger phrase
The phrase that starts the behaviour, like Run Risk Review or Create Exec Summary.
Inputs
The information the AI needs before it acts: audience, goal, files, context, rules, data or constraints.
Output format
The shape of the answer: table, checklist, executive brief, Markdown, JSON, decision log or action plan.
Guardrails
Rules that stop the AI inventing facts, overreaching, ignoring risk or wandering off into jazz flute territory.
Pass criteria
How you know the answer is good enough. This is the difference between “looks nice” and “actually useful”.
Test command
A simple way to test the skill after installing it, so you know it fires correctly.
Failure modes
The common ways the skill can go wrong, plus what to do when it does.
Compatibility
Where the skill works: ChatGPT, Claude, Gemini, local models, automation tools, MCP or human-reviewed workflows.
When should you use a skill?
Use a skill when the same kind of AI work keeps coming back and you want it done consistently.
Use a skill when…
- You repeat the same prompt over and over.
- You need a consistent tone, format or process.
- You need the AI to follow safety or quality rules.
- You want a workflow other people can reuse.
- You want the output tested before it is trusted.
Do not overbuild when…
- You only need a quick one-off answer.
- The task changes completely every time.
- You do not know the outcome yet and are still exploring.
- The instruction is shorter than the setup.
- You are building complexity because it feels clever. Sneaky trap, that one.
A simple skill example
Here is what a practical skill can look like. Not magic. Just clear instructions, useful boundaries and a test.
Skill name: Meeting Action Extractor Trigger: When I say "Extract Actions", review the meeting notes and produce a clean action list. Inputs needed: - meeting notes - attendees - project or team name - deadline context, if known Output format: Markdown table with: Owner | Action | Deadline | Dependency | Risk | Follow-up question Rules: - Do not invent owners or deadlines. - If the notes are unclear, write "Unclear" and ask a follow-up question. - Separate decisions from actions. - Flag anything that looks blocked. Test command: Extract Actions from these notes: [paste notes] Pass criteria: The output should contain only actions supported by the notes, with unclear items flagged instead of guessed.
How to build one
Start small. A skill does not need to be a novel. It needs to be clear enough that the AI can repeat the behaviour without you babysitting it every time.
Name the job
Describe the repeatable task in plain language. Example: “turn messy notes into a decision-ready summary”.
Define the trigger
Pick the phrase that activates the skill. Keep it memorable and hard to confuse with normal chat.
List the inputs
Tell the user what the AI needs before it can do the job properly.
Lock the output shape
Give the AI a specific format so the result is easy to read, reuse and judge.
Add guardrails and tests
Tell the AI what not to do, then give it a test command and pass criteria.
Where to go next
Once the idea clicks, use the hub pages to install, test and combine skills into larger workflows.
Skill Guide
Go deeper into skill structure, installation, platforms and testing.
Open guideBrowse Skills
Explore working examples and copy the ones that match your use case.
Browse skillsUse Case Finder
Start with the outcome and let the hub point you to the right format.
Find my use caseNeed a skill built for your workflow?
Describe what you are trying to do and get routed toward a prompt, skill, workflow pack, AI team, quality gate or automation-ready setup.