Asset Reviewer - Subjective Quality Review Plugin

Status: Experimental

Asset Reviewer is a Claude Code plugin that performs subjective quality review of LLM assets in the otc-awesome-llm repository. It complements the automated CI checks (schema validation, quality scoring) by evaluating dimensions that require judgment: coherence, actionability, cross-platform consistency, and fitness for purpose.

Quick Start

Session-Only (Testing)

claude --plugin-dir claude-code/plugins/asset-reviewer

Permanent Installation

# Add marketplace (if not already added)
claude plugin marketplace add /path/to/otc-awesome-llm

# Install
claude plugin install asset-reviewer@otc-awesome-llm

# Verify
claude plugin list

Commands

/asset-reviewer:review

Reviews locally changed asset files on your current branch.

# Review files changed vs main
/asset-reviewer:review

# Review all assets in the repository
/asset-reviewer:review --all

What it does:

1. Detects changed files via git diff against main
2. Classifies files by asset type (agent, skill, prompt, plugin, etc.)
3. Routes to specialist reviewers sequentially
4. Produces a structured report with pass/fail/warnings per asset

/asset-reviewer:review-pr <number>

Reviews asset files changed in a specific GitHub PR.

# Review assets in PR #299
/asset-reviewer:review-pr 299

What it does:

1. Fetches changed files from the PR via gh pr diff
2. Same specialist review pipeline as local review
3. Displays the report locally
4. Asks if you want to post the report as a PR comment

Architecture

Command (review or review-pr)
  |
  v
Orchestrator
  |-- Detect/fetch changed files
  |-- Classify by asset type
  |-- Route to specialists sequentially
  |
  v  Sequential Specialist Execution
  |
  1. Agent Reviewer     → VS Code agents, Claude agents, shared agents
  2. Skill Reviewer     → Codex skills, shared skills
  3. Prompt Reviewer    → Shared prompts, VS Code instructions, collections
  4. Plugin Reviewer    → Claude Code plugins
  5. Cross-Platform     → Consistency across IDEs (runs last, sees all findings)
  |
  v
Structured Markdown Report

Specialists run sequentially so each can build on the findings of previous specialists. The Cross-Platform Reviewer runs last and has access to all accumulated findings.

Specialists

Agent Reviewer

Reviews agent files for: purpose clarity, instruction coherence, tool alignment, scope appropriateness, actionability, and safety guardrails.

Skill Reviewer

Reviews Codex skill files for: role definition, workflow completeness, domain accuracy, constraint clarity, example quality, and tool integration.

Prompt Reviewer

Reviews prompts, instructions, and collections for: task fitness, instruction precision, output guidance, context sufficiency, tag relevance, and lifecycle appropriateness.

Plugin Reviewer

Reviews Claude Code plugins for: architecture clarity, command design, agent coordination, instruction quality, error handling, and consistency with established patterns.

Cross-Platform Reviewer

Reviews all changed assets for cross-IDE consistency: naming conventions, coverage gaps between IDEs, metadata alignment, and pattern adherence across the repository.

Report Format

Each specialist scores assets on 6 criteria (1-5 scale):

• PASS: All criteria score 3 or above
• WARN: Any criterion scores 2
• FAIL: Any criterion scores 1

The final report includes a summary table, per-asset findings, cross-platform notes, and prioritized recommendations.

When to Use

• Before submitting a PR with new or modified assets
• During PR review to get a structured quality assessment
• After CI passes to catch subjective issues that automated checks miss
• Periodically with --all to audit the full asset library

Related

• scripts/validate-schemas.js — Automated schema validation (CI)
• scripts/quality-check.js — Automated quality scoring (CI)
• tests/integration/asset-integrity.test.js — Integration tests (CI)
• claude-code/plugins/drzero/ — Dr. Zero autonomous improvement plugin