A skill is how your agent does a kind of task: the steps, what the result looks like and the rules to follow. For example, "Rewrite for social media" spells out how to write a Xiaohongshu note, an X thread or a WeChat article, and "Earnings quick read" takes every number from the source. Connections decide which apps your agent can reach; skills decide how a task is done well. They are kept apart.
Skills carry no permissions of their own: sending to other people, deleting and running commands on your computer still ask you first. Using a skill costs nothing extra; the turn spends credits for the model as usual.
| Skill | What it does |
|---|---|
| Rewrite for social media | Turns a brief, article or research into an X thread, Xiaohongshu note, WeChat article or LinkedIn post |
| Research report | Sharpens the question, chooses between your material and deep research, puts the answer first and sources every fact |
| Competitor weekly | Sums up a week of moves by a few companies from your watches and library, ranked by impact, with what it means for you |
| Write a newsletter issue | Turns a period of your watches into an issue ready to send |
| Video and podcast notes | Key points by timestamp, notable quotes, the people and companies mentioned |
| Paper quick read | Problem, method, results, limitations and significance, with a link to the paper |
| Earnings quick read | Headline numbers, year-over-year and quarter-over-quarter changes, guidance and management commentary, numbers only from the source |
| Meeting notes to actions | Decisions and action items from meeting notes, or a brief for a meeting on your calendar; works better with Granola and Google Calendar connected |
To drop one you don't want, open it and select Hide: your agent stops using it and it leaves the / menu. Show again brings it back.
Save your own way of doing things as a skill, and your agent follows it next time:
Open one of your skills on the skills page to:
You can have up to 50 skills of your own, each up to 64 KB. Uploading skill files and importing from GitHub come later.
Skills follow the open Agent Skills standard: a SKILL.md file that starts with a name and a description, followed by the steps in Markdown, optionally with reference files. Skills you used in other tools can come along later.
Developers can list, read, create, change and delete skills with /api/v1/skills, and pass skill (the skill's slug) when sending a message to use one. See the API reference.