Skills

Curated, on-demand instructions and workflows the agent can load mid-conversation.

Skills are bite-sized, self-contained units of agent expertise. Each skill is a folder containing a SKILL.md (instructions + frontmatter) plus any supporting files — references, scripts, templates — and lives in your browser's virtual workspace under /skills/<name>/.

When the user's request matches a skill's trigger description, the agent loads that skill's instructions on demand instead of carrying every workflow in its system prompt.

OpenBrowse implements the open agentskills.io standard, so any skill authored for that ecosystem — whether shared across other agents or installed from the skills.sh community registry — works in OpenBrowse without modification.

Skills are stored in OPFS (origin-private filesystem), local to your browser profile. They never leave your machine.

Bundled skills

OpenBrowse ships with three skills out of the box:

Skill	What it does
`python-env`	Pyodide runtime reference for `executePython` — pre-built package list, browser-env substitutions (e.g. `pdftoppm` → `get_pixmap`, no `subprocess`), `pyfetch` idioms, common pitfalls (top-level `return`, latin-1 PDF unicode errors). The agent loads this whenever it's about to write non-trivial Python.
`find-skills`	Search the skills.sh registry for additional skills and install them via `install_skill`.
`writing-skills`	Author a new skill for the user. Activates whenever the user asks to "make a skill", "turn this into a skill", or "save this workflow as a skill" — drafts a well-formed `SKILL.md` and installs it via `create_skill`.

Invoking skills in chat

Type / in the side panel input to open the slash-command picker. It lists every enabled skill with its trigger description; selecting one prepends a hint that nudges the agent to load that skill via the skill tool before continuing.

The agent can also invoke skills autonomously — when it sees a task that matches an enabled skill's description, it'll load it without being asked.

Managing skills

Open Settings → Skills for the full skill manager:

File tree (left pane). Every installed skill appears as a collapsible folder. Click any file to view it in the right pane.
Viewer (right pane). Markdown files render with preview/code toggle and line numbers. Code files render with Shiki syntax highlighting.
Global enable/disable toggle. Turn the entire skill system on or off without uninstalling anything.
Per-skill toggle. Disable individual skills you want to keep around but not surface to the agent.

Importing skills

Skills can be imported in three ways:

From a .zip or .skill archive. Drag-and-drop the file onto the Skills tab. OpenBrowse extracts it (via fflate) and validates that the archive contains a SKILL.md with frontmatter.
From a URL or GitHub repo. Ask the agent: "install the react-best-practices skill from <url>". The agent uses the install_skill tool.
By authoring one in chat. Ask the agent to write a new skill for you. Once you approve the draft, it persists via the create_skill tool.

The find-skills bundled skill makes search-then-install a one-step workflow:

"Find a skill that can help me write better Tailwind components."

Try in chat

The Skills tab's "Try in chat" button on each skill prefills the side panel chat with a starter prompt for that skill and switches focus to the side panel window. Useful for trying out a freshly installed skill without retyping its purpose.

Tools the agent uses

The skill subsystem is exposed to the agent through three tools:

Tool	Description
`skill`	Load a specific skill's `SKILL.md` into the conversation.
`install_skill`	Install a skill from a GitHub repository or URL.
`create_skill`	Author and persist a new skill directly into the local registry.

See the Tools Reference for full signatures.

Authoring a skill

A minimal skill is a folder with one file:

my-skill/
└── SKILL.md

SKILL.md must start with YAML frontmatter containing name and description. The description is the trigger — write it as a plain-English statement of when the skill should activate.

---
name: my-skill
description: Use whenever the user asks to convert a CSV to a markdown table.
---

# my-skill

1. Parse the CSV with `pandas` (already auto-loaded in `python-env`).
2. Format the result as a markdown table.
3. Return the table inline.

Larger skills can include supporting files alongside SKILL.md — reference docs, scripts, templates. The agent can read any of them with the Read tool using the skill's path (e.g. Read({ file_path: "/skills/<name>/references/<file>" })).

You don't have to write skills by hand. Ask the agent to author one for you — "make a skill that converts CSVs to markdown tables" — and the bundled writing-skills skill will draft a well-formed SKILL.md, walk you through the trigger and scope, and install it via create_skill once you approve.