Show and tell: an auditable, probe-driven code-review skill

[kyy-logs]

I've been working on a different take on code-review skills and wanted to share it here.

The problem I kept hitting: an AI review can produce a confident report, but it's hard to tell what was actually checked. This skill separates the review into four independent probes:

• behavior
• tests
• repository rules
• impact

Each probe writes its own JSON. The human report is rendered from those records, and the merge gate is derived from JSON rather than from the model's prose. It is static-first and read-only.

I like the Standards/Spec separation in this repo; this is a different cut: four operational probes plus a machine-checkable gate.

It's beta, and deliberately not a replacement for tests, CI, or human review. I'd especially value feedback on whether the four-probe split is useful and where the gate is too strict.

Repo: https://github.com/kyy-logs/code-review-skill
Release: https://github.com/kyy-logs/code-review-skill/releases/tag/v0.1.0-beta.3