Question 1

How do I define an agent in Ellipsis?

Accepted Answer

Write one YAML file under agents/ in a repository Ellipsis is installed on. A top-level ellipsis: block marks the file as an agent; the rest declares the instructions, model, trigger, permissions, sandbox environment, skills, and budget. Only the ellipsis: block is required, and every other field has a default.

Question 2

How do agents deploy?

Accepted Answer

By merging. The version of the file on your default branch is the live agent: a new file registers the agent, an edit updates it, a delete removes it, and a rename keeps the same agent with its history and schedules. There is no deploy step and nothing to restart.

Question 3

Can I test an agent before it goes live?

Accepted Answer

Yes, two ways. Open a pull request that edits the file, and sessions triggered by that PR run the branch version of the config. For agents a PR will not trigger, run the file directly with agent session start --config-file, which runs it as written without touching the deployed agent.

Question 4

What happens if I push a broken config?

Accepted Answer

It never deploys. The last good version keeps running, the exact validation error is recorded on the agent in the dashboard, and a pull request that breaks a config gets a comment from Ellipsis naming the error, so it is caught in review, not after merge.

Question 5

What are skills and how do agents get them?

Accepted Answer

A skill is a directory with a SKILL.md: instructions the agent loads by name when the task calls for them. Skills in a cloned repository under .claude/skills/ load automatically at the session checkout, and the config skills list attaches more from other repositories, including one shared skills repository serving every agent in the org.

Question 6

Can agents run automatically?

Accepted Answer

Yes. An agent declares at most one trigger: a cron schedule, a react trigger on pull requests, pushes, issues, Linear issues, Sentry alerts, or Slack events, or an @ellipsis mention route. An agent with no trigger still runs on demand from the CLI, API, or dashboard.

Reusable coding agents

Define once, run repeatedly

Deploy with a merge

Choose when agents run

Frequently asked questions

Explore the platform

Agents

Sessions

Environments

Handlers

Secrets

Webhooks

Models

Observability

Security

Budgets