Lesson 3 · 18 min
Every Monday, someone checks the dashboard, copies numbers, and writes a summary. It’s ideal boring work: read-only, every week, and checked in two minutes. Have a website? The dashboard is Search Console, Google’s free tool that shows how people find your website. Don’t have one? Use the sheet for this week’s numbers. You export the data, and the agent reads and writes.
Start the lesson
Goal: turn the Search Console export (or your weekly numbers) into a report of up to 250 words with checked numbers.
Click each step. You do the first and last.
The same process works for any dashboard that exports a spreadsheet: sales, scheduling, customer service.
In Search Console, using your login. Click each step.
Sign in to Search Console with your account, choose the site, and open the Performance report (search results).
It shows clicks, impressions, CTR (clicks divided by impressions), and average position.
In the date filter, choose to compare the last 7 days with the previous period.
That way, the table shows both weeks side by side, and the agent doesn’t have to guess the dates.
Click Export and choose Google Sheets, Excel, or CSV (a spreadsheet in plain text).
The file has one tab for each dimension (queries, pages, countries…). For the report, you only need queries and pages. Tab names change with your account language.
When the dashboard doesn’t have a value, it shows ~ or -; in the exported spreadsheet, those cases become zero.
In chat, attach the file to the Project and say "report". If the file is large, paste only the first 50 rows from each tab.
In the terminal, save the file in the agent’s folder and run the command in the block below.
No Search Console? Use the sample data: it’s made up, but formatted like a real export.
Query | Clicks this week | Clicks previous week custom cabinets denver | 48 | 30 custom kitchen price | 22 | 41 custom furniture denver | 64 | 60 custom wardrobe | 17 | 9 custom tv panel | 12 | 15 custom home office | 9 | 4 small custom closet | 6 | 13 downtown furniture store | 21 | 21
report Use the sheet in this Project. The data is above (or in the attached file). Before writing the report, show me the total clicks for this week and last week so I can check them.
Role: You are my weekly numbers agent. About me: [what the business does]; I paste the numbers from each tool every Friday. When: when I paste the numbers and say "numbers". Steps: 1) make a table: metric, this week, previous week, change in %; 2) write 3 short bullets: the best news, the biggest concern, a question for the team. Rules: read only, never change or delete anything; save everything as a draft, I send it; never ask for a password, login screen = stop and call me; never pay, approve, or schedule a payment; copy the numbers exactly; if a number is missing, write "couldn't find", never estimate; if you get stuck, stop and ask. Delivery: the table and the 3 bullets, nothing else. First task: run it now with the numbers I’m going to paste.
cd my-agents/site-report # the sheet is in AGENTS.md; the exported spreadsheet is in this folder codex exec --skip-git-repo-check -m gpt-6-luna -s read-only -o report.md "report: follow AGENTS.md with the spreadsheet in this folder"
cd my-agents/site-report # the sheet is in CLAUDE.md; the exported spreadsheet is in this folder claude -p --model sonnet "report: follow CLAUDE.md with the spreadsheet in this folder" > report.md
These are the totals calculated by hand from the sample data. The agent’s wording may vary; the numbers must not.
The agent doesn’t need to access Search Console: it gets the spreadsheet. That means no password in the conversation, nothing changes in the dashboard, and the only possible mistake is incorrect wording, which you’ll catch when you check it.
Run the agent with the sample data and paste its report here. The page looks for the right numbers: the total for both weeks and the biggest drop. If any are missing, the agent got the math wrong or skipped what matters.
Start with the sample data, then use your own.
Create a Project called "Website report" using the example sheet from lesson 2 (the website report one), paste in the example data and Monday’s request. Paste the report into the checker.
Done when: the checker responds OK: the report includes 199, 193, and the drop from 41 to 22.
Export your week from Search Console (or gather your week's numbers) and run the same prompt. No website? Use the weekly numbers sheet.
Done when: you found two numbers from the report in the spreadsheet, and both matched.
Read the report and note one thing you’d like to change (order, length, one more number). Add a line to the sheet.
Done when: the sheet has a new rule, and the next report already reflects the change.
The explanation appears after your answer.
Question 1 of 4
On the chat path, who logs in to Search Console?
Question 2 of 4
Using the sample data, what’s the total number of clicks this week?
Question 3 of 4
The report is well written, but you don’t have time to check it. In the first few weeks, the minimum is:
Question 4 of 4
One of your weekly numbers doesn’t appear in the spreadsheet you pasted. What does the sheet tell the agent to write?
Choose an option. The explanation appears after you try.