vstorm-code-review

Invoke the skill on the pull request after merge conflicts are resolved. It runs Scope, Correctness, Security, and Quality finder subagents in parallel, then Verification and a Judge, and posts one GitHub review with a summary plus an inline comment per raised finding.

Re-review anchors to the commit SHA stamped in the prior review, reads existing comments to reconstruct finding statuses, and scans only the diff since that baseline. Fixed findings get a resolution note, unfixed ones are re-raised at full severity, and new defects in fix commits are raised.

Yes, pass a model argument such as claude-sonnet-4-5 when invoking the skill. Every subagent in the pipeline, from Scope through the Judge, is spawned on that model's tier so the whole review grades consistently.

No, the skill only proposes a GitHub issue destination for pre-existing or unfixed minor findings in its output. A human files the issue from the proposal; review agents never create or close GitHub issues autonomously.

Only verified findings gate the merge, so a possibly hallucinated blocker cannot freeze it. An unverified blocking finding is still raised but behaves as major until concrete evidence, such as a runtime reproduction or airtight static trace, promotes it.

Do not run it while the pull request still has unresolved merge conflicts. It is also unnecessary for negligible or unreachable issues, which the taxonomy explicitly routes to no destination rather than raising them.