A prompt is a single instruction you run once. A skill is a reusable file that teaches your AI how to do one kind of task every time. An agent is an AI set up to carry out multi-step work on its own, often using skills and tools to get there. Prompt = one ask. Skill = standing know-how. Agent = an autonomous doer.
Updated 24 September 2026. Prices and product names read from the live store on the day of writing; the product page is always the authority if they ever drift apart.
These three power most of what people do with AI today, and mixing them up leads to buying the wrong thing. Here is the clean distinction, when to use each, and how they fit together.
The thirty-second test for the word "agent"
Before the definitions, one test settles most of the argument about what "agent" means - and it is worth applying to any vendor, this one included.
We would rather say that plainly than sell you the stronger word. A scheduled summary email gets called an agent. A three-step automation gets called an agent. The test above tells you which of the two you are being offered, so you can judge the price against what the thing actually does rather than against what it is called.
Quick definitions
- Prompt - text you type to get a result right now. Powerful, but you supply it fresh each time. A prompt pack is a curated set of these for a topic.
- Skill - a reusable instruction file your AI loads when a task matches, so it performs like a specialist without re-prompting.
- Agent - an AI configured to pursue a goal across multiple steps: it plans, acts, checks and iterates, often calling skills and external tools along the way.
SKILL.md document inside. What differs is how that document is written. A skill's document describes a role and a menu of things it can do, so you stay in charge of the conversation. An agent's document describes a short fixed intake - a handful of questions it asks before it will produce anything - and one named output packet, the same shape every time you run it.
Side by side
| Prompt | Skill | Agent | |
|---|---|---|---|
| What it is | A single instruction | Standing know-how (a file) | An intake plus a fixed deliverable |
| Lasts | One message | Across every chat | Across one whole task |
| You do | Write it each time | Install once | Answer its questions |
| Best for | Quick one-offs | Repeatable tasks | Multi-step workflows |
| Effort | Lowest | Low, one-time setup | Higher, more capable |
| Right when | You can state the whole job in one sentence | The job recurs and you want the same standards | The job has steps you would rather not sequence yourself |
| Fails at | Drift across a long chat | Quietly not loading when your phrasing misses | Carrying an early mistake through every later step |
| Price here | From $9 | $19 | $39 |
A one-off need, like writing a single article, is exactly what a prompt pack is built for: no setup, just paste and go each time you need a new piece.
View AI Article Writer →One job, three formats: an Amazon KDP launch
Say you are launching a book on Amazon KDP. The same problem looks different under each format:
- Prompt pack. A set of proven prompts for the book description, the category choice and the author bio. You paste them into whichever AI you already use, once, for this launch.
- Skill. A royalty-and-pricing specialist you load once and reuse for every future pricing decision on every book.
- Agent. You hand over the manuscript details, answer its intake questions, and get one finished listing packet back: description, keywords, categories and a pricing recommendation.
The deciding question: will you face this exact kind of decision again next month? If yes, a skill. If it is a single task today, a prompt pack. If it has several steps and you want the finished output rather than parts to assemble, an agent.
What agents do not do, and why that is fine
Because "agent" gets used loosely across the industry, it is worth being precise about what an agent file on this store actually is: a fixed intake followed by one compiled deliverable, run in a single chat session with you answering the questions. That is genuinely useful, but it is not the same as an autonomous system running unattended, and the honest limits are worth stating plainly.
- An agent file does not browse the internet on its own. If Claude has web access in your setup, it has it either way - the file does not add it.
- An agent file does not connect to your CRM, bank, repository or ad account. No credentials, no connection layer. It works with what you type or paste into the conversation.
- An agent file does not run while you sleep. There is no background process - one session, in a chat, with you in it answering the intake questions.
- An agent is not smarter than a skill, it is more disciplined. It removes your choices on purpose so the deliverable comes out the same shape every time, whether you are the one driving it or a colleague on their first week is.
Asks for your product idea, dimensions, landed cost, competitor listings, target price and ad budget, then returns one FBA Launch Packet: a go or no-go scorecard, a margin waterfall down to profit per unit, full listing copy, and a 14-day advertising ramp plan.
View Rhea →That structure is also how you can spot an agent listing before you buy: it names its deliverable ("FBA Launch Packet", "Client Acquisition Packet") and tells you upfront what it is going to ask for. A skill listing describes a role instead and leaves the shape of each answer open.
When to use each
Use a prompt, or a prompt pack, when...
You need a result now and the task varies each time - brainstorming, a one-off email, rewriting a paragraph. A prompt pack saves you from reinventing good prompts for a recurring topic.
Worth understanding why a prompt's usefulness has a shelf life within a chat. A prompt lives inside the conversation - it is a message, and the moment you send it, it starts competing with everything that comes after. Ask a few follow-ups, paste a document, request three variations: by message fifteen or twenty, your original brief is buried under newer, narrower material, and the answers drift back toward generic. The model has not forgotten your instructions; they are just competing for attention with something more recent and more specific. A skill sidesteps this because it is not read from the history at all - it is loaded fresh from a file whenever a request matches, so it is exactly as loud on message forty as it was on message one. More on the mechanism in why long chats get dumber and context windows explained.
Use a skill when...
You do the same kind of task repeatedly and want consistent output without re-explaining - code review to your standards, weekly reports in your format, on-brand copy. Four signals are worth checking if you are unsure:
- The task repeats - anything you will do more than about ten times.
- Consistency matters - three people should not produce three different formats for the same document.
- The role has a method, not just an output - an order of steps, rather than a single instruction.
- You retype the same clarifications every session.
Two or more of those and you have outgrown prompting for that job. Browse skill files.
Instead of re-explaining your SEO standards every time, install Serge once and every future chat gets keyword clusters, technical audits and content briefs done his way.
View Serge →Use an agent when...
The job has multiple steps you would rather not babysit - research then draft then format, or a recurring operational process where the shape of the output should never vary. Browse AI agents.
Rather than running the prompt and the skill yourself step by step, Alistair takes an intake on keyword intent and audience, then returns the structure and the complete article in one session.
View Alistair →How they fit together
They are layers, not rivals. A prompt is the smallest unit. A skill packages know-how so you stop re-typing prompts. An agent orchestrates the work, reaching for skills and external tools via MCP to complete a goal. In practice a capable agent uses skills; a skill replaces a prompt you would otherwise paste every time.
If you already have a prompt that works well, turning it into a skill is mostly mechanical: separate the sentences that say when this applies from the sentences that say how the work gets done. The first becomes the skill's description, the trigger the model matches against. The second becomes the body. The specific input stays out of the file and in your message, every time. Walkthrough here: prompt pack vs skill file.
Which should you buy?
- Just exploring or on a budget - start with a prompt pack, from $9.
- Same task, over and over - get a skill at $19; it pays back fastest.
- A whole workflow with a fixed deliverable - get an agent at $39.
How to tell which one a listing actually is
Two questions cut through most of the ambiguity on a product page. Is there a named deliverable - "Launch Packet", "Client Acquisition Packet" - or a described role? A named packet means agent; a described role means skill. And does the description mention an intake list, specific inputs the file will ask for before it produces anything? That gating behaviour is an agent trait; skills do not usually work that way.
Not sure yet? The free browser tools run the same kind of files with nothing to install and no account, which is the cheapest way to see the "set it once, reuse forever" payoff before spending anything.
How do you install one?
A skill is a ZIP with SKILL.md at the root. In the Claude desktop app, open Customize → Skills, upload the ZIP and toggle it on; it then loads by itself whenever a matching task appears.
Two things gate that menu. The one that catches most people is code execution, switched on as Code execution and file creation under Settings → Capabilities. Plan availability is the other, and Anthropic's own pages disagree: its help centre currently lists Skills on Free, Pro, Max, Team and Enterprise, while its Academy tutorial lists Pro, Max, Team and Enterprise - so if you are on the free plan, check Settings → Capabilities for your own account rather than taking either page's word for it. On Team and Enterprise an owner controls that switch for the whole organisation, which is why the menu is sometimes greyed out on a work account.
Where there is no Skills menu, or in ChatGPT, Gemini or Copilot, paste the contents of SKILL.md into custom instructions instead. You lose the automatic loading; the method still works. Prompt packs need no installation at all.
In summary
For a one-off article, grab the AI Article Writer prompt pack ($9). For consistent SEO work in every chat, install Serge ($19). If you want the whole article taken from intake to finished draft in one session, that is Alistair ($39). All work in Claude, ChatGPT, Gemini or Copilot, with no coding and a 30-day money-back guarantee.
Pick your layer
Match the tool to the job: prompts for one-offs, skills for repeatable work, agents for full workflows. New to the idea? Start with what Claude Skills are.
Related Guides
Pick the layer, not the label
Over 2,000 skills at $19, 300+ prompt packs from $9, and 100+ agents at $39 - each one a readable file you own, not a subscription. 30-day money-back guarantee.
Browse all skills →KissMySkills is a marketplace of 2,000+ AI skills, 300+ prompt packs, 100+ agents & free tools for Claude, ChatGPT & any AI chat.
