The short answers
to common questions.
A quick guide to the catalog, the install flow, community signals, and the checks worth making before you give a skill to an agent.
What is Skilly?
What are skills?
Skills are reusable instruction packages that teach an AI agent how to handle a specific kind of work. They usually contain a SKILL.md with a clear trigger, workflow, expected output, and safety boundaries. A skill is guidance the agent can follow, not model training or a guarantee that the work is correct.
What does Skilly index?
Skilly indexes public agent skill sources, including SKILL.md frontmatter and the full instructions used to make search useful. Every detail page links back to its source.
Does Skilly write or maintain the skills?
No. Skills are maintained by their publishers. Skilly provides discovery, search, source links, install guidance, and catalog metadata.
How do I add a skill?
Where should I run the install command?
Run it from the project directory where your agent harness works. This keeps the skill close to the code and context it is intended to support.
Which agents support skills?
Support depends on the harness. Skilly maintains a practical harness directory; check the harness’s official documentation for the exact loading behavior.
Why cannot my agent see a newly installed skill?
Restart the session or reopen the project. Some harnesses discover skills only when the project starts.
What do the counts mean?
How are skills ranked?
The directory combines catalog install data with anonymous community votes for its popular, hot, and trending views. The exact view is labeled wherever it appears.
Are votes endorsements?
No. Votes are a lightweight community signal, not a security review, quality guarantee, or claim from the skill’s publisher.
How can I report a problem with a skill?
Open the source repository from the skill detail page and report the problem to its maintainers. For an issue with Skilly’s index, use contact.
Is every skill safe?
No. Skilly links to and indexes third-party content. Read the full instructions, inspect the source repository, understand the tools and permissions involved, and start with a bounded project.