Skills & automations

Skills

A skill is a reusable capability the Codex agent can use. Each workspace controls which skills are active for its agent.

Skills come from three places:

System skills — baked into the runner image and always available. They cover platform fundamentals like the JJ workflow and safe terminal usage.
Public skills — curated, reusable skills you can enable from a gallery.
Custom skills — your own scripts, instructions, and policies, scoped to a workspace.

Activation modes

Skills can be set to both, chat, or tasks activation, controlling whether they're available in interactive chat, in scheduled task runs, or both.

Skill triggers

Skills can declare matching triggers — keywords, file patterns, or explicit selectors — that tell the agent when the skill is relevant for a given session.

Automations

An automation runs a saved agent prompt on a trigger:

Manual — fired from the workbench by clicking Run.
Schedule — cron-style schedule. Useful for nightly dependency upgrades, weekly housekeeping, or queue-drain tasks.
Webhook integrations are separate from automations: they receive external events from GitHub, GitLab, Bitbucket, Linear, or Sentry and can create tasks for the workspace.

Each automation run is a chat thread. The thread is auditable in the workspace's automation history — you can see what the agent did, the terminal output, the commits, and the resulting pull request (if any).

Workspace governance

Workspace owners and admins decide which skills are active and which automations exist. Skill activation is opt-in per workspace; the agent cannot reach beyond the skills you've authorised.