How to discover and use AI Skills

Introduction

Choose a concrete task and try one skill on a small, verifiable example before adding it to your workflow.

Install your agent

Install your chosen agent from its official website. This directory links to skills; the agent and source author define installation paths and compatibility.

System requirements

Check the operating system, runtime, network access and dependencies listed by the author. Validate the setup in a separate test project.

Quick start

Search for a task, inspect the source, follow your agent’s installation instructions, then verify an example input. Record the version and configuration.

What are skills?

Skills package instructions and supporting resources for repeatable tasks. Use the detail page to assess relevance, then read the source to understand execution.

Find a skill

Search by task and narrow by category. Compare author, documentation, synchronization date and examples alongside repository popularity.

Install a skill

Follow the source instructions. Claude Code supports project skills in .claude/skills/; OpenClaw users can use its documented Skills CLI. Tool-specific extensions vary.

Use a skill

Provide the goal, inputs and expected output. In Claude Code, /skill-name invokes a skill explicitly; consult your agent for other invocation methods.

Manage skills

Track installed sources and versions. Review changes and test updates before applying them to daily work. Follow the agent’s removal instructions when disabling a skill.

Skill structure

Keep core instructions in SKILL.md. Organize longer reference material and examples by purpose, with clear paths and accessible resources.

my-skill/
├── SKILL.md
├── references/
└── scripts/

Create your first skill

Choose a frequent, well-defined task. Describe its trigger, inputs, steps and result; validate the procedure before packaging it.

SKILL.md metadata

Start with YAML name and description, followed by Markdown instructions. Describe the task and when to use it; consult the Agent Skills specification for field constraints.

---
name: draft-check
description: Review draft text before publishing.
---
# Draft check
Review the provided text and list suggested corrections.

Inputs and parameters

Document required inputs, optional values, defaults and missing-input handling. Parameter syntax is agent-specific; test it in the intended tool.

Test locally

Test normal, missing-input and edge cases against expected results. Record failures and external dependencies, then revise and verify the instructions.

Publish and submit

Publish a repository with a README, examples, version notes and license information. Suggest it through our submission form for directory review.

Environment variables

List variable names, purposes and whether they are required. Use placeholder values in examples and manage real credentials through your tool’s configuration.

Custom tools

Describe commands, scripts or services, including inputs, outputs and failures. Configure authorization in the agent and test each dependency.

Hooks and automation

Hooks are agent-level automation. Check support in the intended tool and define triggers, scope and failure handling before enabling a hook.

Compose skills

Split a workflow into verifiable stages with explicit inputs and outputs. Validate handoffs to avoid conflicting configuration or repeated actions.

CLI reference

These are OpenClaw’s documented lookup and check commands. TopRank does not provide a universal installation CLI; availability depends on your agent version.

openclaw skills list
openclaw skills check
openclaw skills search "pdf"

API and SDK

The directory API and SDK are not released. Register interest; source authors document authentication, limits and failures for any external API a skill uses.

Configuration

Separate personal, project and team configuration. Verify the active scope with a small task after changes and revert one change at a time if needed.

Permissions and data

Inspect requested file, network and execution access. Test with sample data and confirm output locations and external services before real use.

Performance

Reduce repeated instructions and split large references into focused files. Compare results, duration and failures on the same examples before retaining a revision.

Debugging

Check paths and filenames, then environment, dependencies, permissions and logs. Keep a reproducible input and version; remove secrets before reporting errors.

Troubleshooting

Adjust keywords and categories for discovery issues. For execution failures check versions and source instructions; report directory discrepancies with a source link.

FAQ

Listing does not imply independent quality evaluation. Costs, offline support, limits and compatibility depend on the skill, services and agent.

How the directory works

The default order uses GitHub repository stars from our indexed data. Stars belong to the source repository and may be shared by multiple skills. They indicate repository interest, not skill installs or an independent quality score.