Skip to content
RHL AIProductsSkills System

BEHAVIORAL INTELLIGENCE

Agent Skills System

Reusable behavioral intelligence and decision heuristics for autonomous coding agents.

SPECIFICATION ANATOMY

# skills/code-reviewer/SKILL.md
---
name: code-reviewer
description: Expert AI code reviewer for strict TypeScript and performance.
version: 1.0.0
tags: [code-review, typescript, architecture, security]
---
## Operating Framework
1. Evaluate diff against 5 core axes (Correctness, Type Safety, Architecture, Security, Performance)
2. Classify feedback into 3 strict tiers (Blocking, Non-Blocking, Nitpick)
3. Synthesize exact line-targeted replacement solutions

PURPOSE

Why it exists

Hardcoding LLM prompt instructions into frontend JavaScript bundles bloats user payload sizes, couples prompts to application build pipelines, and makes prompt updates tedious.

The RHL AI Skills System decouples intelligence from code. Skills live in declarative markdown schemas with standardized YAML frontmatter, resulting in 0 KB bundle overhead and universal portability across AI agent runtimes.

METHODOLOGY

How it works

01 / ENCODE

Standardized Rules

Engineering domain expertise is written in structured, unambiguous decision trees and negative constraints.

02 / MOUNT

Zero Runtime Bundling

Installed as plain files into your agent’s skills folder (~/.claude/skills/ for Claude Code, or ~/.agents/skills/), then called with /skill-name. Nothing is added to your app build.

03 / ENFORCE

Consistent Execution

Agents follow precise 3-tier severity classification and actionable code diff generation on every task.

VERIFIED SKILLS

Available in the catalog

View all skills →

Code Reviewer

v1.0.0

Expert AI code reviewer enforcing strict TypeScript, 5-axis inspections, and multi-tier feedback.

Frontend Performance Engineer

v1.0.0

Evidence-led web performance diagnostics for Core Web Vitals (LCP, INP, CLS) and bundle optimization.

PR Description Writer

v1.0.0

Rigorous PR descriptions backed by real verification runs and annotated before/after browser evidence.

Install a skill into your agent today

$npx @rhl-ai/cli skill install code-reviewer --target ~/.claude/skills

Then start a new Claude Code session and type /code-reviewer.