Automations
Scheduled tasks for the agent: schedules, run permissions, manual runs, pause, history. A service on the machine runs them — even with the window closed.
An automation is a task for the agent that starts by itself on a schedule: “every Monday at 9:00 run the tests and write down what failed”, “on Fridays put together a summary of the week’s commits”. Each run is a new session in the project; the result waits for you in the session list.
The Buff Code service on this machine starts them — even when the window is closed. If the machine is off, the run at that time does not happen.
The list
The tabs at the top filter automations by state: All, Now running, Awaiting permission, With errors, Paused. Each card has:
- the name and the state — “Active” or “Paused”;
- the schedule and the project: “Mon at 09:00 · teashop”;
- the next run and how many runs there have been; the result of the last one — “Completed”, “Failed”…;
- buttons:
| Button | What it does |
|---|---|
| Run now | an extra run; the schedule does not shift |
| Open session | the session of the last run |
| Pause / Resume | stop the schedule (“Paused — does not run on schedule”) and bring it back |
| Edit | open the editor |
| History | all runs: when, manually or on schedule, how long it ran, the result, a link to the session |
| Delete | delete the automation; the run sessions stay |
A new automation
“New automation” opens the editor:
| Field | What it sets |
|---|---|
| Name | what the automation is called in the list and in the run sessions |
| Schedule | “Every N minutes or hours”, “Daily”, “On days of the week”, “Every N weeks”, “Monthly”; under it — the time and a “Next run — …” line, so you see right away what you have set |
| What the agent should do | the whole task. Each run is a new session: the agent does not see past conversations, so write as if explaining for the first time |
| Project | the folder the agent works in |
| Permissions in a run | “Ask”, “Edit on its own” or “Full access” — the same modes. If a run still asks for permission, it waits for your answer in its session, and the card goes to the “Awaiting permission” tab |
| Model | an organisation model or “Project default” |
The time is the local time of the machine the service runs on.
Run history
An entry can be deleted — the run’s session stays.
How it is recorded
A run session is an ordinary development session: it is visible in the column and in the Buff workspace and goes into the work records just like yours. Usage is charged to the organisation’s plan.
In the console
/automations — the same list with search: schedule, project and next run; choosing a row gives
a manual run, pause, history. /automation-new — a new automation step by step. The agent can do it too —
ask in the chat: “create an automation: check the tests every day at 9:00”.
Git in the window
The Git window in Buff Code: what changed, a commit with a message from the agent, push, branches, history. The branch chip shows the state and opens it.
Workflows
Workflows: subagents work in phases with checks and a result. How to ask the agent, follow a run, save it and run it again with arguments.