caveman-review

Generate one-line, location-specific code review comments with severity markers.

Updated Mar 11, 2026

Search Tags:#diff analysis #code review #pull requests #automated feedback #one-line comments #severity labels

Ready to run

npx skills add https://github.com/selfagency/agentsy --skill caveman-review-selfagency

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: caveman-review
Source: https://github.com/selfagency/agentsy/tree/main/.agents/skills/caveman-review
Command: npx skills add https://github.com/selfagency/agentsy --skill caveman-review-selfagency

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Caveman-review eliminates noisy, verbose pull request feedback by producing compact, paste-ready comments that preserve the actionable signal.

Core Features & Use Cases

Ultra-compressed review formatting: Outputs one finding per line in a strict L<line>: <severity> <problem>. <fix>. style.
Severity-aware guidance: Uses 🔴 bug, 🟡 risk, 🔵 nit, and ❓ q to distinguish incidents, fragility, style, and questions.
Noise reduction for PRs: Drops throat-clearing, hedging, restating the diff, and redundant approval language while keeping exact line numbers and symbol names.

Quick Start

Use caveman-review when you want concise PR feedback by prompting an AI with: "Review this PR using caveman-review format and comment with one actionable line per finding."

Frequently Asked Questions about caveman-review

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

FAQPage Schema

How do I write concise code review comments for a pull request diff?▼

Concise code review comments use a strict one-line format like `L<line>: <severity> <problem>. <fix>.` to directly state the issue and solution. This eliminates verbose feedback while preserving exact line numbers and actionable signal.

What is the best format for automated PR feedback that reduces noise?▼

The best format for automated PR feedback drops hedging and approval language, outputting one finding per line with severity labels like `🔴 bug` or `🟡 risk`. It directly states the problem and concrete fix without restating the diff.

How do I categorize pull request findings by severity in code review?▼

Categorize pull request findings using severity labels: `🔴 bug` for incidents, `🟡 risk` for fragility, `🔵 nit` for style, and `❓ q` for questions. This distinguishes critical issues from minor style nits in one-line comments.

Can I review multi-file PR changes with one-line actionable comments?▼

Yes, reviewing multi-file PR changes with one-line comments works by using the `<file>:L<line>:` structure. This preserves exact line numbers and backticked symbols across single and multiple file diffs while maintaining consistent formatting.

What should one-line code review comments avoid including?▼

One-line code review comments should avoid code writing, automated linting, approval language, and noisy phrasing like throat-clearing or hedging. They must strictly state the problem and concrete fix without restating the diff context.

Does the caveman-review format work for both single-file and multi-file diffs?▼

The caveman-review format works for both single-file and multi-file diffs by adapting its structure. Single files use `L<line>:` while multi-file changes use `<file>:L<line>:` to preserve exact locations across the entire PR diff.