Audit a code boundary

Use audit to inspect a named part of the repository and receive independently checked findings. A boundary can be a file, directory, or subsystem. Without one, the audit covers the current repository.

Examples

Paste either request into your coding-agent chat:

Use zdev to audit hello.py in the context of area greeting.
Check argument parsing and preservation of the default greeting.

Use zdev to audit the command-line interface for correctness and usability.
Use separate review lenses for those two concerns.

What happens

The agent assigns a fresh reviewer to the boundary. Someone other than the author of a substantive finding checks its location and claimed consequence before it reaches the final report. You can request up to four explicit review lenses, such as correctness and security; these use separate reviewers.

The report begins with one of these results:

Result	Meaning
PASS zdev-audit	The checked evidence produced no findings.
FINDINGS zdev-audit	The report contains supported findings with locations and impact.
BLOCKER zdev-audit	A required review or evidence check could not be completed.

Expect the boundary, inspected and omitted material, and checked evidence to be named. An audit ends at this report. Repository files, Git state, and task status stay unchanged; fixing findings is a separate request.

For a general list of promising work, use improve. For acceptance of the current ready task, use verify. See the workflow reference for review standards.