Skills

Skills are installable capabilities that extend what an agent can do. The Skills page (Settings → Skills, at /skills) is a directory: browse and install skills, then manage which agents use them. Skills are also surfaced under Connectors → Browse skills.

The skills directory

The directory has two sections:

• Installed — skills already added to this workspace, including custom (chat, upload, URL, or GitHub) skills.
• Available skills — entries from the public ClawHub catalog you can install with one click, plus integration connectors (Telegram, Slack, Discord) that show a Connect button.

A search box filters by title, slug, author, or description.

Where skills come from

Use Add skill to install from any of these four sources:

• ClawHub — paste a skill page URL from the public marketplace.
• GitHub repo — install from a repository (optionally a path within it).
• From URL — install from a hosted SKILL.md or ZIP link.
• Upload file — add a .md, .js, or .zip from your machine.

You can also save a chat as a reusable skill — that's a separate flow started from the Save as a reusable skill button inside a chat, not an Add skill source. See Custom skills.

See Installing skills for the step-by-step flow.

Enabling on agents

Installed skills turn on for every agent by default. Open a skill's card to manage it — the panel has an Overview (agents using it, enabled count, and any required credentials), a Files tab showing the raw SKILL.md source, and per-agent toggles.

Some skills need credentials or variables:

• Credentials / secrets — bind an existing secret (or create one inline) per agent when you enable the skill.
• Variables — pick a secret, type a custom value inline, or create a new secret.

To remove a skill entirely, use Uninstall from its manage panel — that takes it off every agent.

Installation status

After you install a skill it moves through installing → ready (or failed if something went wrong). A skill that needs secrets on some agents shows a Setup needed badge, and a failed catalog install offers a Retry.

In this section