Verzeta

Skills

A skill is an installable instruction pack that an agent can discover and apply. Skills add knowledge or step-by-step instructions to agents without changing their system prompts. Skills never grant tools and never run scripts on their own.

What is a skill?

A skill is a folder with:

Example minimal SKILL.md:

---
name: cli-quickstart
description: How to scaffold a small Python CLI app using typer or click.
tags: [python, cli]
---

# CLI Quickstart

When the user asks to design or scaffold a Python CLI:

1. Choose `typer` for new code (type-hint native).
2. Define one entry function per subcommand.
3. Persist to JSON, one record per file under `~/.<app>/`.

The agent reads this skill when relevant and follows the instructions inside.

Installing skills

You can install skills from two sources:

1. ClawHub (the public skill catalog)

ClawHub is a public skill catalog. To install from it:

  1. Open Skills in the navigation bar.
  2. Open the Search & Install tab.
  3. Search by name or topic and press Enter.
  4. Click Install. Items the catalog flags as malware cannot be installed; suspicious items show the catalog's verdict.
  5. Review and approve the skill (see below).

2. A local folder (skills you write yourself or get from a colleague)

  1. Open Skills in the navigation bar.
  2. Click Import Folder and select the skill's folder. Verzeta copies it into its own skills folder; your original folder is not changed.
  3. Review and approve the skill (see below).

Warning: Skills are user-approved before agents can use them. Every new skill starts as Unreviewed, and agents cannot use it until you approve it. You can also Block a skill to keep it unusable. This is the security boundary: Verzeta will not silently follow instructions from a skill you have not reviewed.

Approving a skill

  1. Open Skills and go to the Review Queue tab, or find the skill under Installed.
  2. Click Review & approve to open the skill's instructions, frontmatter and any warnings.
  3. Read it. Tick I understand that I am responsible for how this skill is used.
  4. Click Approve, or Block to reject it.

The skill moves to the approved state. Agents can now discover and use it.

Pro-tip: Treat skill approval the same way you treat installing a new browser extension. Read what it does before you click Approve.

How agents discover skills

If a chat or its project has preferred skills, each request lists them with a one-line description. Skills that declare the tools they need are tagged:

If there is no preferred list, agents are told to call discover_skills to see the approved skills.

The agent can then call:

The agent uses what it learns to drive its next reply.

Preferred skills: pointing agents at the right skills

Every approved skill is available in every conversation through discover_skills. To make agents use particular skills, set a preferred list:

Preferred skills are listed at the start of every request, in the order you set. Tick Use / Expose Only Preferred Skills to hide all other skills from agents in that scope.

Skill order matters

Preferred skills appear in the order you set, with ready skills before blocked ones. If two skills conflict, put the one you want followed first.

Removing a skill

On the Skills page, find the skill and click Remove skill. This deletes the skill's files and removes it from every preferred list. To use it again you must install it again.

Authoring a skill

To create a skill of your own:

  1. Make a folder named after the skill (e.g. cli-quickstart/).
  2. Inside, create SKILL.md with YAML frontmatter (name, description, optional tags and tools) and the instructions in the body. Verzeta builds the skill's id from the name (or the folder name), using lowercase letters, digits, hyphens and underscores.
  3. (Optional) Add other files (templates, examples, reference docs).
  4. On the Skills page, click Import Folder and select it.
  5. Review and approve it.

Pro-tip: Keep the description to one clear line, because that is what agents see in every request. Keep the body short too: an agent reads the whole body when it opens the skill, and it counts against the context window.

What's next

Verzeta Studio guides