Skills
A skill is a folder of instructions an AI reads when the work calls for it - a SKILL.md and supporting files, in the Agent Skills open standard the wider ecosystem uses. Holochain development, debugging, writing, internal comms: a skill is know-how, not a program.
Getting one
Add-ons › Skills shows Ready to add - open-source skills we have read and pinned to the version we read (knowledge only, permissive licenses, sized for a local model), plus anything people have shared. One tap adds it. Add a skill takes your own: a folder, a zip, or a link (a GitHub repository or a folder inside one, pinned to today's commit).
A skill that ships scripts, hooks or MCP servers is marked ships programs - the app never runs anything just by adding a skill.
Giving it to an AI
A skill does nothing until an AI carries it: "Used by" on the card, or the Skills section when you edit the AI. Chosen per AI, on purpose - a few chosen skills stay cheap, and not every AI should know everything.
How an AI uses it
- In chat the AI carries a one-line list of its skills and the full text of the one that fits the question, picked on your device by meaning. The size on each card is what that costs when it is used, and the model details on a reply show which skills were carried.
- In a project the agent sees every installed skill's description and reads one only when the work calls for it - "Using skill" shows as a step.
Updates
A skill shows Update available when its source has moved on: a skill added from Add-ons when its listing moves to a newer version, a skill added from a link when its branch does. One tap brings it to the new version. Nothing updates on its own. A listing claimed by its maker reads Signed by @username on its card.
Sharing a skill of your own: Sharing add-ons.