Docs
Knowledge

Skills

Browse OXYGEN's agent skills and write workspace skills that your team's agents can find and follow.

A skill is task guidance an agent reads before it works: a SKILL.md with a name and a description of when to use it, plus any supporting files. A skill grants nothing on its own. It adds no tool, spend, or approval authority, so every paid or external step it describes still asks for its own preview and approval.

In the web app

Open Knowledge → Skills (/knowledge/skills). It lists two kinds of skill:

  • OXYGEN's skills: the product skills available to your workspace, read-only and browsable file by file.
  • Workspace skills: skills your team wrote or imported. Each one is a folder of ordinary Knowledge files under Workspace skills, so its files open in the Knowledge editor and keep their revision history.

Workspace skills

Create one from a name, a description, and its instructions:

oxygen skills workspace create --name follow-up-rules \
  --description "Use when drafting a follow-up to a prospect who went quiet" \
  --instructions "Wait five business days. Reference the last message. Ask one question." \
  --json

The name is lowercase with hyphens and is stored as workspace-<name>, here workspace-follow-up-rules.

Or import a skill you already have: one SKILL.md, a folder with its supporting text files, or a public HTTPS link to a SKILL.md:

oxygen skills workspace import --directory ./skills/follow-up-rules --json
oxygen skills workspace import --url https://example.com/skills/follow-up-rules/SKILL.md --json

An import is all or nothing and never overwrites an existing skill. It takes up to 50 text files and 1,000,000 bytes in total, at most 120,000 bytes per file. Content is kept verbatim with its source link. Scripts in a skill are stored as text and never run.

Find, read, edit, and retire skills:

oxygen skills workspace list --query follow-up --json
oxygen skills workspace get --name workspace-follow-up-rules --json
oxygen skills workspace update --name workspace-follow-up-rules \
  --content "$(cat SKILL.md)" --expected-revision 2 --json
oxygen skills workspace archive --name workspace-follow-up-rules --expected-revision 3 --json

update replaces one file, SKILL.md unless you pass --file, at the revision you last read; a stale revision is refused. A SKILL.md must keep its YAML frontmatter, because its description is what agents search. archive takes the skill out of discovery and keeps its files and history. None of these commands spends credits or calls a provider.

Workspace skills are shared with every member of the workspace. An agent connected over MCP finds them with oxygen_workspace_skills_list and reads them with oxygen_workspace_skills_get; oxygen_workspace_skills_create, oxygen_workspace_skills_import, oxygen_workspace_skills_update, and oxygen_workspace_skills_archive match the commands above. Where the Workspace Copilot is available, it sees an index of your workspace skills and loads one when a task calls for it.

OXYGEN's skills in your own agent

To give a local agent OXYGEN's product skills, install them with oxygen skills install; the Claude Code, Codex, and Cursor pages show the command for each agent. Over MCP, oxygen_skills_get reads one product skill or one of its files.

  • Knowledge Graph: the store, editor, and revision history workspace skills live in.
  • MCP reference: every oxygen_workspace_skills_* tool and its schema.

On this page