Skills

A skill is a folder with a SKILL.md in it — instructions the agent reads when a task calls for them, and files beside it that cost nothing until they are opened. The Skills screen is the library those folders live in.

The Skills catalog with community and Splox skills — The catalog: skills you can fork into your own library

Three tabs: Catalog is what you can copy, System skills is what the platform ships, Your skills is what you own.

What ships

The System skills tab listing agent-browser, memory, skill-authoring and splox — The four system skills, available in every run

Skill	What it teaches
`agent-browser`	Driving a real browser: reading a page as an accessibility snapshot, clicking and filling, and what to do about bot protection and captchas.
`memory`	Searching everything the user has said across all their chats, and keeping notes that survive the conversation.
`skill-authoring`	Writing a skill that actually fires: the frontmatter rules and how a description gets a skill selected.
`splox`	Building on Splox itself: editing a harness’s programs, hooks and tools, and running evals against a version.

Each has Fork — take a copy into your own library and change it — and Turn off.

Getting one from the catalog

Click a card and you get the skill’s own files, read-only, with its license:

A catalog skill open in the read-only viewer with a Fork and save button — A catalog skill before you fork it: the files, the license, and the SKILL.md

Fork & save copies it into Your skills under a name of your own — it proposes <name>-copy — and from then on it is a file you can edit. Two other ways to fill that tab:

Create skill opens an editor on a fresh SKILL.md, with the frontmatter and the headings already in place.
Import from GitHub takes a link to a skill folder in any public repository, or a directory holding several of them, and pulls them in.

Attaching one to an agent

Skills are attached in code, in the agent’s declaration:

assistant = agent(
    "Assistant",
    system_prompt=prompt("assistant"),
    tools=["system:compute"],
    skills=["system:memory", "system:agent-browser", "trading-desk"],
)

system:<name> is one of the platform’s; a bare name is one of yours. Every skill is listed by name and there is no wildcard, so an agent has the skills its line names and no others.

The library’s files are in the sandbox either way, at ~/skills. The skills= list is what the agent is told it has — the name and the description sit in its context, and the body is read when the skill fires. That is the whole trick: a harness can carry a hundred pages of procedure and pay for a line of it per turn.

Writing a skill

What goes in SKILL.md, what goes beside it, and when a skill beats a prompt.

Agents

Everything else an agent declaration carries.

Start here

Ask your agent

Look inside

Concepts

In the app

Account

Help

What ships

Getting one from the catalog

Attaching one to an agent

Writing a skill

Agents

​What ships

​Getting one from the catalog

​Attaching one to an agent

Writing a skill

Agents

What ships

Getting one from the catalog

Attaching one to an agent