best-practices-reviewer_skill

Overview

This skill checks whether an implementation follows Claude Code best practices and prompts targeted questions or advice when improvements are needed. It activates automatically in risky or ambiguous development moments and can be invoked manually as an implementation review. The goal is to reduce wasted work, surface missing verification, and keep sessions healthy and focused.

How this skill works

The skill inspects the current task, session length, and recent edit patterns to detect risky conditions (cross-file changes, ambiguous prompts, lack of tests, repeated fixes). It asks concise diagnostic questions and suggests concrete actions: add verification, sketch a plan, narrow scope, use a sub-agent, or clear session history. When triggered, it recommends specific next steps and offers prompt templates for reworking the task.

When to use it

• Before starting large, cross-file changes
• During long sessions that load many files
• When a prompt or request is vague or underspecified
• If no tests or verification methods exist for the change
• When the same fix has been attempted multiple times
• For ad-hoc checks: “implementation review”, “best-practices check”, or “progress check”

Best practices

• Always define verification: tests, sample inputs/outputs, or screenshots before coding
• Follow explore → plan → implement for non-trivial changes; skip for tiny fixes
• Ask for target files, expected behavior, and edge cases when prompts are ambiguous
• Detect and isolate session contamination; run /clear after repeated failures
• Use sub-agents to survey many files so the main context stays focused

Example use cases

• A developer begins a refactor touching multiple modules — skill prompts for a plan and tests
• A long conversation accumulates unrelated attempts — skill recommends clearing and restating the goal
• A vague request like “make this faster” — skill asks for target files, performance metric, and inputs
• Repeated fixes to the same bug — skill suggests clearing context and redesigning the prompt
• Exploratory tasks: skill confirms whether a loose prompt is intentional and how results will be validated