Agent Skills

An Agent Skill is a reusable folder of instructions and supporting files for a particular task. Its SKILL.md explains when and how the agent should use it. Some skills also contain scripts, examples, or reference material. A compatible AI agent reads these files to guide its work; the folder is not a standalone application.

A prompt gives an AI assistant instructions in a conversation. A Skill packages a repeatable workflow with reusable instructions and, when needed, supporting files. Use a prompt for a quick request; consider a Skill when you repeatedly need the same process or reference material. You still provide the goal and context for each task.

Search for the result you want, such as copywriting, competitor research, or web testing. Compare the tasks, prerequisites, and expected output on the detail page. Start with one skill that matches a small real task and check whether your agent has the file, network, or execution capabilities it needs.

Open a skill's detail page and download its ZIP from GitHub. Keep the complete folder, including SKILL.md, references, scripts, and license. Add it using your agent's skill installation method, complete the listed prerequisites, and confirm that the agent can discover it. Then send the example request with your own materials and check the expected result.

Extract the skill folder into .claude/skills/<name>/ inside your project. SKILL.md should be directly inside that folder, rather than inside another nested copy of it. Use the original skill name when asking Claude Code to apply it. Install any runtime or external service dependencies listed on the skill's detail page separately.

Place the extracted skill folder in .agents/skills/<name>/ inside your project, with SKILL.md directly inside it. Confirm that Codex has discovered the skill before starting a task. A downloaded folder does not automatically configure external tools, accounts, or runtimes; follow the prerequisites for the particular skill.

If custom Skills are available for your account and organization, upload the ZIP through Customize → Skills and enable it. Keep the skill folder and supporting files together in the archive. Skills that depend on local software or unavailable connections may need a different environment even when the upload succeeds.

Check the folder location, an extra nested directory, and whether SKILL.md is present. Confirm that the agent has refreshed its skill discovery, restarting the session if your tool requires it. If the skill is recognized but the task fails, check its runtime, network access, external connections, and required input files. The ZIP download alone does not satisfy these dependencies.

You can ask an agent that supports Chinese to work with Chinese materials and produce Chinese output. Specify the desired language, audience, and terminology in your request. Results depend on the agent and skill, so check the output against your task requirements.

The files in this collection are available from the public open-source-skills repository and its GitHub Releases, with original author attribution and licenses preserved. Read the included license for the particular skill. Downloading is free; the AI tool or external services you use may charge separately.

Compare the source version on the detail page with your installed copy. Save your own changes before replacing the folder with a newer complete package, then repeat a small task to check it. For a manual installation, remove the specific folder from the agent's skills directory. If you used an installer, use its update or removal process.

The current collection has source, license, and package-content checks. It does not claim that every skill has completed real tasks in every agent. Review the files and prerequisites before enabling a skill, and check facts, citations, generated code, and proposed actions in its output. The examples describe expected results rather than verified demonstrations.