An automated status report is only as honest as its prompt

The factory built in public at software-factory.dev writes its own weekly summary for the people following the build. The automation doing the writing is worth opening up for what is missing from it.

It is the Weekly Recap. Once a week it tells the audience what the factory shipped.

What the automation reads and writes

The trigger is a cron line: 0 9 * * 1. Monday, 9am, after the week has closed.

It reads four things:

• seven days of daily metrics files
• last week’s recap, for comparison
• every pull request merged that week, title and description
• every issue moved to done

Then it writes a recap with a highlights section, a numbers table, the features shipped in user terms, what is planned next, and a section called challenges and learnings.

That last section is the interesting one. It is not in there by accident.

Honesty written into the prompt

The one detail worth copying is that the honesty is engineered into the prompt itself. The automation is told, in writing, that the challenges section is important for authenticity, that the language stays factual with no marketing in it, and that it uses real numbers only. Where the human and agent time was never recorded, it has to write “not yet recorded” instead of a figure that sounds right.

Those are four short lines in a config file, and every one of them is doing heavy work. A system that reports on itself drifts toward public relations unless the honesty is written down. Left to default, a self-report rounds up, drops the messy parts, and fills the gaps with something plausible.

An automation is as honest as its instructions

This is the part teams underestimate when they automate a status update. The automation will be as honest as its instructions force it to be, and no more. “Write a recap” produces a brochure. “Write a recap, the challenges section is what makes it authentic, never invent a number” produces something a reader can trust.

You can read the full definition in the weekly recap automation on GitHub (opens in a new tab).

What does your team’s status update do with the week that went badly, and with the number nobody actually measured?

Further reading

• Prompt Engineering →
• Agentic Software Factory: Advanced Team Training →