review-lens_skill

Overview

This skill finds what human code reviewers miss — logical errors, silent edge cases, performance cliffs, implicit assumptions, and state bugs that make code incorrect despite appearing to work. It inspects code through seven specialized lenses to surface failure modes that show up only under real-world conditions. The goal is to catch correctness, safety, and scalability issues before they reach production.

How this skill works

The skill analyzes code using seven focused lenses: Boundary Analysis, Failure Path Analysis, State Transition Analysis, Implicit Assumption Analysis, Performance Cliff Analysis, Security Surface Analysis, and Correctness Under Change. For each lens it runs a checklist of concrete checks (empty inputs, error propagation, N+1 patterns, TOCTOU gaps, etc.) and produces actionable findings with severity, location, and suggested fixes. Output is presented as a concise review summary highlighting critical, high, and medium issues plus passes.

When to use it

• Before opening a pull request to catch deep logical issues early
• When reviewing unfamiliar code or large diffs to focus attention
• After changes that touch money, auth, or persistent state
• When code 'works in tests' but may fail under production scale
• If you have a hunch there’s a subtle bug you can’t reproduce

Best practices

• Run the lens on the branch locally before requesting human review
• Treat findings as bug reports: include reproduction hints and minimal fixes
• Prioritize fixes by severity (critical state and failure-path issues first)
• Use the lens output to augment, not replace, a human review
• Iterate: re-run after applying fixes to validate the remediation

Example use cases

• Detecting order state machine gaps that could charge customers without fulfillment
• Finding N+1 DB queries introduced by a recent refactor before load testing
• Uncovering swallowed errors that return ambiguous None/undefined to callers
• Identifying implicit encoding, uniqueness, or size assumptions before deployment
• Spotting performance cliffs where a function degrades from milliseconds to minutes