ToShop DocsYour First Task
A real walkthrough — give ToShop a goal and watch a full task run end to end.
This walkthrough takes you through every stage of a real task so you can see how your agent thinks, asks for permission, and reports back.
The goal
"Take a screenshot of the frontmost window, extract any text in it with OCR, and save a Markdown summary to ~/Desktop/."
This task chains a few capabilities together: take a screenshot, run OCR on it, write the result to a file, and send you a desktop notification when done.
Walkthrough
Prompt
Open Chat and paste the goal. Your agent confirms it understood, then drafts a short plan:
- Capture the frontmost window.
- Run OCR on the image.
- Summarize the extracted text.
- Write
~/Desktop/summary-{timestamp}.md. - Notify when done.
Approvals
Your agent pauses for approval on:
- First file write to a new path under
~/Desktop/— you can approve once or always. - First time taking a screenshot in this session — same choice.
Approvals you grant become rules in Settings → Permissions and persist across sessions until you revoke them.
Execution
Your agent runs the steps in order, streaming each tool's result back into the chat.
If a step fails (for example, OCR finds no text), your agent asks how to proceed instead of guessing — you stay in control.
Result
You see the screenshot, the OCR text, and a link to the Markdown file. The notification appears. Done.
What just happened
The planner
Broke the goal into steps before doing anything. The plan is visible in chat — you could've edited or rejected before any tool was called.
The permissions system
Gated every system-touching action. Read-only ops ran automatically; writes (screenshot, file write) asked first.
Memory
Recorded the approvals you gave so the next similar task is smoother.