Skill

An AI skill is a method an AI can reuse for a kind of work. A prompt asks for one result and a tool gives it an action, while a skill says what to collect, which rules to follow, what resources to use, and how to check the result.

For a business, that means a proposal skill can bring the same intake questions, house template, approval rules, and checks to every proposal. But you still have to agree on the method before the AI can repeat it.

The open Agent Skills format requires a SKILL.md file with a name and description, and it can include instructions, scripts, references, and templates. Agents load that material in three layers, starting with a short catalog entry, then the full instructions, then any supporting files the job needs.

What does an AI skill contain?

A skill puts the instructions for one kind of work in one place so an agent can find and reuse them. In the open Agent Skills specification, each skill is a folder with a required SKILL.md file. That file starts with a name and description, followed by the work instructions. The folder can also contain scripts, reference material, templates, and other assets.

That structure means you don’t have to put every company rule into every request. The Agent Skills overview describes three levels, the agent sees the name and description first, reads the full instructions when the task matches, and opens supporting files only when it needs them.

For proposal writing, the main instructions can set the sequence, collect the client’s problem, confirm the scope, choose the right template, draft the proposal, and run the approval checks. A reference file can hold current service descriptions, a script can check dates, owners, and required clauses, and a template can carry the approved layout.

The skill gives you a place to keep the method, but the method still has to work. If scope lives in one manager’s memory and three partners use different approval rules, putting their notes in a folder doesn’t settle the disagreement.

How is a skill different from a prompt, tool, or workflow?

These four parts solve different problems.

A prompt gives instructions for one request, such as “turn these notes into a project brief,” and needs to be supplied again for the next request. A tool lets the AI read the CRM or create a document, without teaching the firm’s method.

A skill teaches it how to create and check project briefs repeatedly, with an agreed standard and an owner. The workflow handles the order from the end of the call through creating, reviewing, and filing the brief, rather than teaching every step itself.

So a skill can contain prompts, call tools, and run inside an AI workflow. Keeping the parts clear helps you find where something went wrong.

If the CRM tool returns the wrong record, check the connection or query, and if risks go in the wrong section, check the skill’s instructions, example, or test. If the brief never reaches the project manager, the workflow handoff needs fixing, and calling all of those problems “the AI” makes it harder to see who should repair them.

A skill also does not replace the workflow. It can teach an agent how to complete a step. The workflow still needs a trigger, context, handoff, and record. A person may still own the decision at the end.

When should a firm turn work into a skill?

Create a skill when the same work repeats and the best version of that work can be written down. Four conditions make the candidate clear.

The task repeats. A one-off board memo can stay a prompt. A project brief created after every discovery call is a candidate for a skill.

The standard is stable. The people responsible for delivery should agree on required inputs, acceptable output, and the failure route. A skill cannot settle a live management dispute.

Real examples exist. Include strong outputs, common edge cases, and known mistakes. Generic examples teach generic work.

Someone owns it. One person or team approves changes, watches failures, and removes stale rules. Shared use without a named owner turns a skill into an old procedure that happens to run.

The usual failure starts small, one employee keeps a prompt in a notes app, another copies it and changes two lines, and a third starts from last month’s output. The same missing field gets repaired by hand, but nobody puts the fix into a shared method, so the team keeps working around a gap in the process.

Start by writing down the work. Document the process first, including judgment calls and exceptions, then package the agreed part as a skill.

That order matters because a skill makes instructions easier to repeat, including instructions that are wrong.

How do you test and maintain an AI skill?

Test the work the skill produces. Do not use the agent’s statement that it followed the instructions as the test.

Anthropic’s October 16, 2025 engineering guide says to start with evaluation, run tasks like the ones you’ll use it for, look at specific failures, and build the skill around the knowledge or procedure that’s missing. That keeps the instructions tied to a problem in the work.

Build a small test set from work the firm already handles. For a proposal skill, include a clean standard request, incomplete notes, conflicting dates, an unsupported promise, and a request outside the service scope. Score whether the skill collects missing information, uses the right template, keeps claims inside the evidence, and routes exceptions to the right person.

Use a written eval so you can compare results even when the model changes, and run the same cases before and after each skill edit. Save the skill version with each result, because fixing one missing field while adding two false promises makes the work worse, even if the proposal reads better.

When your firm changes a service, template, approval rule, or tool, the skill needs to catch up even if nobody has edited its folder yet. Its owner updates the files, reruns the tests, and records the new version, the same care any shared process needs.

What can go wrong with an AI skill?

The first risk is making a bad process repeatable. A detailed skill can look mature while carrying the wrong approval rule into every job. More instructions can make that error harder to notice.

The second risk is scope. A single “do everything” skill loads rules that do not belong together. Proposal writing, contract review, invoicing, and client onboarding have different owners and permissions. Split them along real responsibility lines. Smaller skills are easier to trigger, test, and retire.

The third risk is access, because skills can contain code to run and instructions to contact outside systems. Anthropic recommends using trusted sources and checking the files, other software they depend on, and outside connections before use, just as you would with downloaded software.

Give each skill the least access its task needs. A proposal skill may read approved service information and create a draft. It does not need permission to send the proposal, change the CRM owner, or read payroll. The decision about autonomy still depends on whether the result can be checked and whether the action can be undone.

An outdated skill creates a quieter risk. The file may still run while the service, template, or approval rule has changed. A named owner and repeatable test are what make that drift visible.

Where does a skill sit on the Delivery Model Ladder?

A production skill belongs at Stage 2, Augmented, on the Delivery Model Ladder. At that stage, AI is part of a named workflow with a trigger, current context, tool access, a review point, and a final record. The skill supplies reusable know-how inside that route.

The same file can sit at Stage 1, Enhanced, when a person manually chooses it, pastes in the material, checks everything, and moves the output by hand. Packaging instructions does not change the delivery model by itself. The surrounding route decides the stage.

This is also how skills fit inside an AI employee, whose role needs a defined scope of work. Skills teach part of the method, the workflow supplies the start, handoffs, limits, and record, and a person still owns the result until measured work supports giving the system more freedom.

Quick answers

Is an AI skill just a long prompt? No. A skill can include prompts, but an agent can find and reuse it when the task matches, along with scripts, references, templates, examples, and rules for when to load them.

Does a skill give an agent new tools? A skill can teach an agent how and when to use tools, and include scripts it can run. The tool gives it an action, while the skill teaches the method around that action.

Who should own an AI skill? The person responsible for the delivery process should own the standard. A technical teammate can maintain the files, but the person who answers for the work should approve process changes.

Are downloaded AI skills safe? Treat them like downloaded software, read the instructions and scripts, check what else they install, what access they need, and which outside systems they contact. Run them with limited access first and test real examples before giving them more access to client data or production systems.