Show and tell: probe-level code review records and a JSON gate

[kyy-logs]

Sharing a code-review skill for people building agent workflows.

The part that may be interesting here is the separation of responsibilities: four independent probes (behavior, tests, repository rules, impact) each write their own JSON, and the coordinator derives the report and merge gate from those records instead of asking one model to summarize everything.

It is a standalone Claude Code / Codex skill, static-first, and read-only. Beta, not a replacement for tests, CI, or human review.

I'd be interested in feedback on whether this kind of probe-level record is useful when composing larger agent workflows.

Repo: https://github.com/kyy-logs/code-review-skill