reviewer

Review only changed files with fresh context and return a structured verdict.

Updated Mar 12, 2026

Search Tags:#workflow-automation #code-review #atlas #fresh-context #changed-files #structured-verdict

Ready to run

npx skills add https://github.com/arvidfcjarfalla-prog/atlas --skill reviewer-arvidfcjarfalla-prog

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: reviewer
Source: https://github.com/arvidfcjarfalla-prog/atlas/tree/main/.claude/skills/reviewer
Command: npx skills add https://github.com/arvidfcjarfalla-prog/atlas --skill reviewer-arvidfcjarfalla-prog

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Fresh-context code review. The reviewer sees ONLY changed files — no accumulated history, no sunk-cost bias.

Core Features & Use Cases

Correctness — Logic errors, off-by-one, null safety
Edge cases — Empty arrays, missing keys, concurrent access
Security — Injection, auth bypass, secret exposure
Backward compatibility — Breaking changes to public APIs or types
Atlas-specific rules — AI never generates raw MapLibre; manifest compiler is the only path; compiler stays dumb

Quick Start

Provide the list of changed files and run the reviewer to obtain a structured verdict.

Frequently Asked Questions about reviewer

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

FAQPage Schema

How do I review only changed files to avoid sunk-cost bias in code review?▼

Fresh-context code review isolates changed files by spawning a sub-agent that sees only those files and your CLAUDE.md rules. This prevents accumulated history from skewing the assessment.

What is a structured verdict in automated code review?▼

A structured verdict is a pass or fail result returned after a sub-agent evaluates changed files. If the verdict is FAIL, it triggers iterative fixes up to three loops before continuing the workflow.

How do I automate iterative code fixes when a review fails?▼

When a fresh-context review returns a FAIL verdict, the workflow automatically triggers iterative fixes for up to three loops. If the review passes, the workflow continues without intervention.

What security and edge case checks should a fresh-context code review cover?▼

Fresh-context code review checks for logic errors, null safety, empty arrays, missing keys, injection vulnerabilities, auth bypass, secret exposure, and breaking changes to public APIs.

Can I use fresh-context code review for Atlas-specific MapLibre rules?▼

Yes, the review enforces Atlas-specific rules such as ensuring AI never generates raw MapLibre and that the manifest compiler remains the only path for compilation.

What do I need to provide to start a fresh-context code review?▼

You need to provide the list of changed files. The reviewer then spawns a sub-agent with a fresh context containing only those files and your CLAUDE.md rules to produce a structured verdict.