Report

Findings

Every finding carries:

• The claim — the observable condition the check states, e.g. “Fetch call does not provide an AbortSignal.”
• The check — doctor/check-id, the rule that fired.
• The location — file:line, and the count when the same shape appears more than once.
• Severity — error, warning, or info.

Checks also explain themselves: impact (what goes wrong), why (the mechanism), and fix (what to do). These fields ride along in the JSON output so your agent reads the same explanations you do.

Blind spots

Blind spots are written by your agent for each doctor. When your agent builds a doctor, it declares what that doctor cannot catch. They are there to guide your agents through the code review process. Think of it as your agent consulting with your doctor, so it can make the educated decisions itself.

Where blind spots come from — and how agents are taught to probe their own — is covered in honest checks.

The score

A 0–100 number at the end of a run. Severity is weighted across findings and files, with a grade:

An empty scan (zero files checked) doesn’t score a perfect 100 — nothing checked is a warning, not a win.

Broken doctors are infrastructure failures

A doctor that fails to load or crashes mid-scan is reported as broken, fails the run outright, and the score renders as null. Broken tooling never reads as a clean codebase.