A skill is a procedure Starko loads only when it applies. You write it once in the knowledge hub, publish it, and the same skill reaches the ⌘L chat, Starko in Slack, your MCP host and Starko AI's coding sessions. It follows the Claude skill format: the body becomes SKILL.md, and attached files become its references/, scripts/ and assets/.
Write it in the knowledge hub
- A new Skills tab next to Articles, with search and a status per skill.
- New skill ("Teach Starko a skill") starts blank or from an internal article. Any internal article's menu also has Turn into a skill. Skills are always internal and are never published on your help center.
- The editor has When to use (what Starko reads to decide whether to load it), a Files card with a folder tree, inline editing and uploads, and a preview of the generated
SKILL.md. - The sidebar holds Available in (⌘L chat, Slack, MCP, Coding sessions), Who can use it (People and Starko, Only people via /, or Only Starko), Projects, Scripts and Versions.
Drafts stay drafts until you publish
Publishing creates an immutable version, and agents only ever read published versions, so editing a draft changes nothing Starko runs. Publishing an unchanged draft creates no new version. Skills with scripts or network access can only be published by an owner or admin.
In ⌘L and Slack
Starko sees the name and "when to use" of each published skill it may pick and loads one when a request matches. You can also call one yourself: type / in the ⌘L composer for a menu of the skills you may use, with a Workspace or project badge on each, or write $name. Money like $20 never opens the menu. A skill you invoke shows up as a chip in the sent message; click it for the version, its instructions and Copy.
Bundled scripts run in a sandbox, after you approve
Starko can run a skill's .sh, .py or .js script from ⌘L or Slack. Every run asks first and names the file, the skill version, the arguments and the network access. Scripts run in an ephemeral sandbox that is deleted afterwards and never sees your secrets. Network is off unless the published version lists domains, and only owners and admins can add them. The result card shows the duration, exit code and output. If the skill is republished between your approval and the run, nothing runs.
In Claude, Cursor and VS Code through MCP
Skills with MCP switched on appear as prompts in your host's own / menu, through the Starko MCP connector. Picking one inserts its SKILL.md, plus your request if you add one. "Only Starko" skills stay out of that menu. Reconnect the connector or start a new conversation to see newly published skills.
In Starko AI coding sessions
Skills with Coding sessions switched on go to fix runs and pairing sessions for the projects you pick (or all projects). The agent finds them in a skills/ folder in the checkout, which is kept out of git so they never land in a pull request. The Publishing section shows when a version is ready for coding sessions. Sandbox setup, project learning and eval replays run without them.