A specialized agent for reviewing, scoring, and providing actionable feedback on AI-generated outputs using a structured rubric.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install critic-agent
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install critic-agent using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Critic Agent Skill?
The Critic Agent Skill acts as an independent quality assurance layer within your AI ecosystem. It evaluates agent outputs against four critical dimensions: Correctness, Clarity, Completeness, and Safety. By providing an objective numeric score and detailed improvement suggestions, it enables developers to implement robust quality control loops and iterative refinement processes in complex multi-agent systems using Openclaw Skills.
This skill is essential for maintaining high standards in automated pipelines, allowing users to define quality gates that must be passed before any output is finalized. Whether you are generating code, technical documentation, or data analysis, the critic ensures that every piece of content meets your specific requirements and safety guidelines.
Critic Agent Skill Use Cases
- Implementing automated quality gates before delivering AI-generated content or code.
- Creating iterative feedback loops where a writer agent refines its output based on critic suggestions.
- Standardizing evaluation criteria across a diverse team of specialized AI agents.
- Auditing large-scale agent outputs for safety compliance and factual accuracy.
How Critic Agent Skill Works
- The skill receives the original task description, the agent's generated output, and any additional context or requirements.
- It applies a weighted rubric to evaluate the output based on Correctness (40%), Clarity (25%), Completeness (25%), and Safety (10%).
- An LLM analyzes the content and assigns individual scores and detailed feedback for each dimension.
- A final weighted score (0-100) is calculated and compared against your configured pass/fail thresholds.
- The system returns a structured JSON response containing the overall score, specific feedback, and actionable suggestions for improvement.
Critic Agent Skill Setup
To start using this skill in your environment, install it via the CLI and run a test critique with the following commands:
# Install the critic agent
openclaw skills install critic-agent
# Run a basic critique
openclaw skills run critic-agent \
--task "Write a Python function to parse CSV files" \
--agent-output "def parse_csv(path): ..." \
--context '{"requirements": ["handle edge cases", "include docstring"]}'
Critic Agent Skill Data Schema & Taxonomy
The Critic Agent Skill evaluates outputs based on a weighted scoring system and returns data in a structured format:
| Dimension | Weight | Criteria |
|---|---|---|
| Correctness | 40% | Factual accuracy, logical soundness, and absence of errors. |
| Clarity | 25% | Readability, organization, and effective communication. |
| Completeness | 25% | Coverage of requirements, edge cases, and thoroughness. |
| Safety | 10% | Ethical considerations, compliance, and lack of harmful content. |
Scores are categorized as Excellent (80-100), Good (70-79), Needs Revision (50-69), or Fail (0-49).
Critic Agent Skill Advanced Features
- Automated Retry Loops: Configure workflows to automatically trigger revisions if the critic score falls below a specific threshold.
- Task-Specific Customization: Tailor the critic focus for code reviews, written content, or data analysis by providing specialized context.
- Multi-Model Support: Run evaluations using high-reasoning models like Claude 3.5 Sonnet through Openclaw Skills integrations.
- Configurable Fallbacks: Define custom behaviors for API timeouts or malformed responses to ensure your pipeline never hangs.
- Integration Hooks: Easily embed the critic into complex multi-agent orchestrations using YAML-based workflow definitions.
SKILL.md
Loading
Related Openclaw Skills
Nika Skill Creator
A specialized framework for engineering, refactoring, and validating AI agent skills compliant with the Nika platform constraints.
News Crawler
An automated news fetching and summarization tool designed to extract content from RSS feeds and websites to generate structured reports.
sense-music
A music perception skill for AI agents to analyze BPM, key, structure, genre, mood, and lyrics from audio files.
Ecommerce Copy Humanizer ZH
A sophisticated tool designed to strip away the robotic tone of AI-generated Chinese e-commerce copy, replacing it with natural, high-conversion human-like expression.
Ecommerce Image Asset Generator
A strategic AI tool for planning and generating high-conversion ecommerce visual assets that align with specific business objectives.
Plex Media Server
A comprehensive toolset for managing Plex libraries and sessions through Openclaw Skills.
Featured*
