Automations

An automation runs the Squasher Agent on a schedule, or when an incident closes. The agent collects its own evidence, starts work on the problems it finds, and tells you only when something needs a person. A run that finds nothing new stays quiet. Automations are separate from the automatic triage and fixes of new issues. To change those, see AI triage and Auto-fix.

​

Create an automation

1. Open your project and select Automations in the sidebar, under Alerting.
2. Select New automation.
3. Under Choose a template, select Create active automation on the template that you want.

The new automation is active and runs on the schedule of its template. Only project owners and admins can create, run, pause, archive, or delete automations.

​

Templates

Select Templates on the Automations page to read what each template does. You can also run every automation by hand.

​

What happens after a run

Most templates run as a read-only Squasher Agent conversation. When the conversation ends, the agent reports one of these results: Each finding also opens a thread in Chat, so you can ask the agent about it. When the agent needs a person, for example to decide on a risky change or to take an action that it cannot do, it asks one question. Squasher sends a high-priority notice to the person on call for the project. When nobody is on call, it goes to the person who created the automation. Answer in the thread that the notice opens. When the agent offers numbered answers, reply with a number or with your own answer. The same question on a later run joins the same thread and does not notify again on the same day. When a finding is a new or worse issue, the agent can start AI triage on it. Triage then follows your project’s automatic triage and Auto-fix settings, so a fix pull request opens only when those settings allow it. The agent also records what it learned in its memory when the lesson matters for later runs. It also reads the results of its earlier runs, so a later run does not report the same cause again. CI Failure Grouper, Dependency Drift Reviewer, Untested Path Finder, and AGENTS.md Workflow Updater produce a structured report instead. Their outputs can need approval.

​

Run, pause, or remove an automation

Open the automation from the Workflows view. On its page:

• Run now starts a run at once.
• Pause stops the schedule. Activate starts it again.
• Archive sets the automation to archived, so it does not run.
• Delete removes it. Confirm with Delete automation.

Recent runs lists the runs. Open a run to see its Steps and its Recorded outputs. An agent run stays Running until the agent submits its report. If the agent stops without a report, or sends no report within three hours, the run fails. Dispatch alone does not count as a successful run. When the project’s daily agent budget is spent, the run fails and the automation runs again at its next trigger. An incident-close run gets the resolved incident and up to 20 of its latest timeline events. A manual run of an incident-close automation fails when there is no incident to read.

​

Limits

• An automation does not open pull requests itself. It starts triage, and Auto-fix opens the pull request when your settings allow it.
• The budget fields of an automation are stored, but Squasher does not stop a run when it goes over each configured field. Separate agent safety limits stop excess model calls and context growth. A run that reaches a safety limit fails and retains its recorded usage.
• Text from logs, replays, and incidents is data for the agent, not instructions.

​

Manage automations with the API

The API paths are under /v1/projects/{project_id}/automations, /v1/projects/{project_id}/automation-templates, /v1/projects/{project_id}/automation-runs, and /v1/projects/{project_id}/automation-capabilities. API keys need these scopes: See Automations API.

​

Next steps