mkl-review-pr

Review pull request diffs for concrete correctness defects with evidence-backed findings.

11|1|Updated Sep 13, 2026
One-click install
npx skills add https://github.com/00200200/maintainer-skills-lab --skill mkl-review-pr-00200200
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: mkl-review-pr
Source: https://github.com/00200200/maintainer-skills-lab/tree/main/providers/cursor/.cursor/skills/mkl-review-pr
Command: npx skills add https://github.com/00200200/maintainer-skills-lab --skill mkl-review-pr-00200200

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Pull request reviews often miss real regressions or drown authors in subjective style comments. This Skill inspects the full diff, surrounding code, and relevant tests to surface only actionable defects with concrete triggers and consequences. ## Core Features & Use Cases - Defect-focused review: Traces callers, data shapes, error paths, and compatibility promises to find bugs with a concrete trigger and user-visible consequence. - ML and data reproducibility checks: Runs static reproducibility scans (e.g., Repro Lens) and bounded replays when the repository documents them, treating RNG, device, and data-order findings as review questions. - Structured findings output: Each finding includes a title, file and line, triggering conditions, consequence, and supporting evidence, with uncertainty labeled and style preferences kept separate. - Use Case: A maintainer asks for a review of a PR changing a training pipeline; the Skill reads the diff, runs the repository's declared replay command, and reports one actionable seed-handling defect with exact reproduction details. ## Quick Start Ask the assistant to review the current pull request for correctness regressions using its diff, surrounding code, and tests.

Frequently Asked Questions about mkl-review-pr

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I review a pull request for correctness bugs?▼

Establish the base and head revisions, read the complete diff plus surrounding code, and trace callers, data shapes, and error paths affected by the change. Report only defects with a concrete trigger and user-visible consequence, each with file, line, and supporting evidence.

How to review ML or data-processing pull requests for reproducibility?▼

Look for the repository's reproducibility checker and declared replay command, run a static scan such as Repro Lens if available, and compare before/after reports. Treat new RNG, device, or data-order findings as review questions rather than automatic proof of a bug.

Can this review skill approve or merge a pull request?▼

No. The Skill does not approve, merge, post comments, or modify the patch unless those actions are explicitly part of the user's request. It only returns findings and states validation limits.

What happens when a pull request review finds no defects?▼

The Skill states that no actionable defect was found and describes the validation limits, including the scope reviewed and checks performed. It does not manufacture findings to fill a quota.

Does a clean static scan prove an ML change is correct?▼

No. A clean static scan or a matching replay run is evidence only for the scope checked, not proof of scientific validity or cross-platform equivalence. Findings are reported with the exact revision, environment, command, and declared outputs.