Automations
Automations start cloud agents for you — on a schedule or when something happens in your workflow (a new pull request, a failing check, a comment). You write the instruction once; X4RGE runs it whenever the trigger fires.
Think of it as putting a recurring or event-driven task on autopilot: the same teammate in the cloud, same repo and setup, without you clicking "Start run" every time.
When to use automations
Use automations when the job is repeatable and well-defined. For one-off exploration, start a manual cloud run in the Agents workspace instead.
In practice
Each run uses your cloud setup — repository, install steps, secrets, and network rules — so automated runs behave like manual ones.
How automations work
Pick a trigger
Choose when the automation runs: a recurring schedule, a GitHub event, or a manual run from a saved template.
Describe the task
Write what the agent should do each time — scope, severity, and what to do when nothing needs changing (for example, "report no issues" vs. stay silent).
Scope repository and branch
Point the automation at a repository (and branch) so the agent clones the right code. Event triggers may pass a specific ref, such as a pull request branch.
Test, then enable
Run once manually, review the output, then turn the automation on. Watch the first few scheduled or event-driven runs in Sentinel before widening scope.
Like Cursor, Devin, and Factory recommend for background agents: try automations on a low-stakes repository first. Confirm prompts and triggers behave as expected before production repos.
Ways to start a run
GitHub must be connected for repository-scoped and webhook-driven automations. Schedule and manual runs still need a repo when the task edits code.
Slack and other sources may appear as trigger options in the UI. For production automations today, prefer GitHub or schedule triggers. See Troubleshooting if a trigger is listed but not yet connectable.
Starter templates
The Agents workspace includes templates you can customize:
Pick a template, adjust the prompt and trigger, and save. Templates are starting points — tune instructions for your team's standards and approval habits.
Create an automation
Open Automations
In the Agents workspace, go to Automations (or start from a template).
Choose a trigger
Schedule (time and timezone), a GitHub event, or manual-only. Confirm the next run time shown in the UI for schedules.
Write the instruction
Be specific: what to inspect, what "done" looks like, and whether to open a PR or only comment. See Prompting for task-writing tips.
Set repository, model, and approvals
Select the repo and branch. Choose a model and how strictly the agent must ask before sensitive actions — stricter for automations that open PRs, lighter for read-only triage.
Enable and monitor
Save, run once by hand, then enable the trigger. Use Sentinel to spot misfires, duplicate PRs, or runaway triggers early.
Best practices
Prompts should behave safely when run more than once. For example: "If a fix PR already exists for this failure, comment on it instead of opening a duplicate."
- Scope narrowly at first — one repo, one event type, clear success criteria.
- Pair with usage limits — use Usage and org budget policy so a noisy trigger cannot exhaust allowance.
- Review before merge — automations propose work; your team still approves what ships.
- Reuse cloud setup — keep install steps and secrets aligned with CI so automated runs don't fail on environment drift.