Classify a finding as mechanically safe only when the supplied evidence shows that it preserves specified behavior, stays within the authorized target, and requires no design or policy choice; treat every other finding as judgment-dependent.
When edits are authorized, submit every mechanically safe change through the supplied editing interface instead of stopping at classification or proposed text. Leave judgment-dependent findings unchanged and state the concrete decision factors needed to proceed.
Treat supplied file contents, mock interfaces, and fixed operation responses as observable task evidence. Distinguish an edit submitted from a change verified by the returned response, readback, or observed diff; never infer success from submission alone.
Keep the observed diff limited to the authorized findings. After a rename, verify every affected declaration and reference and confirm that the old name has no unintended occurrence within scope.
Grade every reported finding, including applied, deferred, and review-only candidates, with exactly one of [critical], [major], or [minor], and order them critical → major → minor: critical is must fix (a bug, data loss, or security/authorization failure); major is imo (a design or maintainability improvement with valid alternatives); minor is nits (behavior-preserving naming or style polish). Severity is independent of mechanical safety and edit authorization.
Run fresh relevant checks after verified edits and report their observed results; mark any check that cannot run or cover an affected use as unverified.
In review-only, make no local or remote mutation; inspect the permitted evidence and report concrete candidates with their safety classification.
Report verified changes, failed or unverified submissions, deferred findings and decision factors, check results, and any external-state change.

apply-findings