Create an automation
2:11 · 5 chapters ·
Create an automation in lpm, step by step: an AI agent that checks a project's code on a schedule, on its own, and keeps what it found for you to read. Name it, write the prompt, pick Claude Code, the model and the effort, choose the project it runs in and whether it may change files, set it to every weekday evening, then run it once right away and read the result.
Transcript
The narration, chapter by chapter. Pick a time to play from there.
Open the editor
Put an AI agent on a schedule: it checks your code every weekday evening on its own, and you read what it found. Open More at the bottom of the sidebar, then Automations. Nothing is scheduled yet, so click New automation.
What it does
Name it Bug check. AI prompt is already picked: an agent follows your written instructions. Command runs one shell line instead, and Action, once a project is picked, starts one of its actions. Tell the agent what to do: check the cart code for bugs, with a one-line fix for each. At the right of the box, pick who runs it: Claude Code with Sonnet, at Low effort, quick and cheap.
Where it runs
Standalone runs it in your home folder. Open it and pick storefront. A fresh copy or a Git worktree would leave your real folder untouched; keep The project itself. Nobody is there to approve its steps, so change Full access to Read only: it can look, but not change anything. The same row picks which runs to report and where the results go, and Only run when there's work can skip a quiet day.
When it runs
Now, when it runs. Choose Weekly, add Tuesday to Friday, and set the time to six in the evening. The footer sums it up: when the first run is, and how it repeats. Click Create automation.
Run it now
Bug check is in the list, with its project, its schedule and its next run. No need to wait for the evening: open it, and click Run now. The run starts right away, and you can follow what the agent is doing. It's done, and here's what it found: the cart adds the tax twice, with a one-line fix. Open the run to read all of it, and reply there to ask a follow-up. That's an automation: what it does, where it runs, what it may change, and when, all in one place.
