Discover
Search by the outcome you need. Skilly indexes names, descriptions, categories, sources, and full instructions so a useful skill does not have to be found by name alone.
SKILLS EXPLAINED What are agent skills?
What are agent skills?
Reusable know-how for agents.
A skill is a small, reviewable instruction package for one repeatable kind of work. It gives an agent context, a workflow, and boundaries it can apply when a request matches.
A skill turns repeatable knowledge into an agent-readable playbook.
When people ask an agent to do a familiar kind of job, the hard part is often the procedure: what to inspect first, which decisions matter, which tools are appropriate, and what a useful result should look like. A skill records that procedure in a form an agent can read and follow.
ReusableUse the same workflow across many projects and conversations.
ReviewableRead the source as plain text before giving it to an agent.
BoundedState assumptions, permissions, risky actions, and when to ask first.
A skill does not retrain the model, install a new model capability, or automatically make an agent correct. It supplies focused instructions and context. The harness decides when to load those instructions and the agent still needs to interpret them.
Most skills start with one file.
The canonical file is usually named SKILL.md. Its YAML frontmatter helps a catalog and an agent understand what the skill is for; the Markdown body explains how to carry out the work.
SKILL.MDMINIMUM EXAMPLE
---
name: accessibility-review
description: Review a web interface for keyboard access, semantics, focus states, and readable contrast.
---
# Accessibility review
## Use this skill when
The user asks for an accessibility review of a web page or component.
## Workflow
1. Inspect the relevant files and identify the page boundaries.
2. Check semantics, keyboard access, focus states, and contrast.
3. Report findings with a concrete fix and a way to verify it.
## Safety
Do not change files until the user asks for implementation.FrontmatterA recognizable name and precise description with useful trigger language.
WorkflowOrdered steps, decisions, examples, and the expected result.
BoundariesTools, permissions, network access, destructive actions, and stop conditions.
Supporting files such as scripts, templates, or reference notes can sit beside SKILL.md when they make the workflow more reliable. Keep the first version small and explain how each extra file is used.
A skill moves through a simple lifecycle.
Inspect
Read the complete instructions, open the source repository, and check what tools, files, network access, or permissions the workflow expects.
Install
Use the publisher’s install command from the project where your agent works. Project scope is a good default because the change stays visible and reversible.
Apply and review
Give the agent a bounded request, review the result, and improve or remove the skill if its instructions are unclear or too broad.
Skills are one layer of an agent system.
| Concept | What it provides | Typical use |
|---|---|---|
| Prompt | Instructions for one request or conversation. | “Review this pull request.” |
| Skill | Reusable instructions for a class of requests. | A repeatable pull-request review workflow. |
| Tool | An executable action available to the agent. | Read a file, run a test, or make an API call. |
| MCP server or integration | A connection that exposes tools or data from another system. | Query an issue tracker or design system. |
A skill can explain when and how to use tools, but it is not itself the tool. Whether a harness supports a particular skill format, directory, or automatic discovery behavior varies; check the harness documentation when installation or loading is unclear.
Good skills make the next action obvious.
- The name and description say what the skill does and when to use it.
- The workflow has a clear start, finish, decision points, and expected output.
- Examples use realistic inputs and do not rely on hidden context.
- Required tools, permissions, network access, and destructive actions are explicit.
- The skill stays focused on one repeatable job instead of becoming a general-purpose instruction dump.
SAFETY FIRST
Read a skill before you run it.
Skills are code-adjacent input from a publisher. Instructions can influence which files an agent reads, which commands it runs, and which services it contacts. Skilly provides discovery and source links; it does not certify every skill or replace your review.
Start with a bounded project, keep credentials out of skill files, grant only the access the task needs, and ask for confirmation before destructive or external actions.
Read the security FAQ