Team skills are enabled per organization. Contact support to turn them on.
Write a skill
Set it up
In the app, open Skills under Agents in Settings and choose the Team tab. Every member can read the team’s skills. Team owners add, change and delete them, and on the default team, organization admins and owners do. Through the API, use an API key for the team:items, next_cursor and actions. actions["team.manage_skills"] contains the policy presentation: state (enabled, hidden, disabled, confirm or refused), reason, message and confirm. Use it to decide whether to show editing controls; the API does not require clients to recompute authorization. Pass each next_cursor unchanged to read the next page.
The single-skill response is the saved skill. Send its version to replace or delete it.
Use it in a run
Runs onxhigh, on scout with graph: true, and on high list the team’s skills. To make every agent of a run follow a skill, attach it with skills on POST /people-intelligence-async. POST /v1/people/intelligence does not take skills yet, so this request uses the original route, which keeps working.
sixtyfour_atlas_start takes the same skills. The Atlas API’s start and continue endpoints don’t.
Limits
Requests to
/v1/skills and the original /skills routes share one rate limit.
Good to know
- A run keeps the list it started with. Reading a listed skill that isn’t attached returns its latest saved text, and one deleted since is refused.
- The frontmatter holds
nameanddescription. Alicenseormetadatafield, as Claude Code and Codex skills carry, is dropped. Any other field is refused. - A skill is one file. Link a full
https://address, not a file beside it. - Saving with an old
versionsaves nothing. Read the skill again and reapply your change. - A run that attaches an unknown skill or too much text is refused before it starts or is charged.
- An API key acts with its creator’s current role in the team. See Roles and Permissions.