caveman-review

Generates one-line code review comments with location, severity, problem, and fix.

Updated Sep 19, 2026
One-click install
npx skills add https://github.com/Ab0umar/selrs.cc.BU --skill caveman-review-ab0umar
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: caveman-review
Source: https://github.com/Ab0umar/selrs.cc.BU/tree/main/.agents/skills/caveman-review
Command: npx skills add https://github.com/Ab0umar/selrs.cc.BU --skill caveman-review-ab0umar

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Code review feedback is often buried in verbose, hedged prose that slows down both reviewers and authors. This Skill compresses PR review comments into single actionable lines, cutting filler phrases while preserving line numbers, symbol names, and concrete fixes. ## Core Features & Use Cases - Terse Comment Format: Outputs findings as L<line>: <severity> <problem>. <fix>. with severity emoji (🔴 bug, 🟡 risk, 🔵 nit, ❓ question). - Noise Filtering: Drops throat-clearing, hedging, per-comment praise, and restatements of what the diff already shows. - Auto-Clarity Mode: Switches to full paragraphs for CVE-class security findings, architectural disagreements, and onboarding contexts, then resumes terse output. - Use Case: Paste a multi-file pull request diff and receive paste-ready review comments like L42: 🔴 bug: user can be null after .find(). Add guard before .email. ## Quick Start Ask the AI to review this pull request diff using caveman-review and return one-line comments for each 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?▼

Use the format L<line>: <severity> <problem>. <fix>. with one line per finding. Prefix with severity emoji (🔴 bug, 🟡 risk, 🔵 nit, ❓ question) and keep exact line numbers, backticked symbols, and a concrete fix instead of vague suggestions.

What is a good format for PR review feedback?▼

A single line per finding works well: location, problem, and fix, such as `L23: 🟡 risk: no retry on 429. Wrap in withBackoff(3).` Drop filler like "I noticed that" and restating what the diff already shows.

When should code review comments be verbose instead of terse?▼

Use full paragraphs for CVE-class security findings that need explanation and references, architectural disagreements requiring rationale, and onboarding contexts where a new author needs the reasoning. Terse one-liners suit routine bugs, risks, and nits.

Can this skill approve or request changes on a pull request?▼

No. It only outputs review comments ready to paste into the PR. It does not approve, request changes, write code fixes, or run linters; those actions remain with the human reviewer.

How do I turn off terse review mode?▼

Say "stop caveman-review" or "normal mode" to revert to verbose review style. The skill also auto-expands individual findings when they involve security, architecture, or onboarding contexts.