Themes
What each bot keeps saying, grouped into themes.
On this page
The Themes tab of Bots Monitoring sums up what your review bots keep raising. AI reads the bots’ comments in the window and groups them into recurring themes, with how serious each theme is and where in the code it clusters. You read a short summary instead of hundreds of comments. The tab shows only while AI is on.
#What you see
The bots mostly flag money handled as floating-point numbers and missing error handling around the payment provider. Most of it is in acme/payments.
Amounts are parsed into floats before rounding, so a refund can be a penny out. 14 comments.
Where: acme/payments · refunds/
Calls to the payment provider catch errors and carry on. 9 comments.
- A summary of the window in a few sentences.
- A severity strip: how the themes split between critical, major, minor and nit.
- Theme cards, each with a name, a description, a severity, a category (such as Correctness, Security, Error handling or Testing), the comment count and Where in the code it appears. Select a theme to see its threads, or a pull request to open it.
- By reviewer: each bot’s comment count and the share of its threads later resolved or likely addressed.
- A coverage line, for example “Summarised 310 distinct points from 512 bot comments”, and when the summary was generated.
#Generate a summary
- Open Bots Monitoring and choose Themes.
- Check the window on the ROI tab. Themes uses the same window.
- Press Generate. The button says Summarising… while it runs.
- Later, press Regenerate for a fresh summary. Comments that have not changed are not read again.
A summary uses your own AI provider (see Claude or Codex) and counts towards your AI usage.
#Why it helps
The numbers on ROI tell you how much a bot says. Themes tells you what it says. A theme that repeats every sprint is a fix worth making once, in the code or in your standards, rather than a comment to answer again on every pull request.