Agent checkpoint or pre-commit gate?

A user asked whether ViewDoctor's diff scan should run after every coding-agent change or only before commit. The old workflow made both interpretations possible.

In 0.1.7 I split them:

Agent checkpoint after one coherent edit:

viewdoctor scan . --git-diff --format agent

Hard boundary after staging:

viewdoctor scan . --staged --fail-on warning

The first includes new untracked Swift files and keeps the response compact. The second reads only the Git index and can block a hook or CI. On a large local fixture the compact response was 95.2% smaller by bytes with the same findings.

My current preference is one checkpoint after a coherent slice, then the staged gate. For teams using coding agents: what cadence has worked without turning feedback into noise?