ClawRank is a comprehensive, evidence-based scoring framework designed to evaluate and track AI agent performance across seven critical dimensions.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install nate-clawrank
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 nate-clawrank 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 ClawRank?
ClawRank provides a structured methodology for assessing the efficacy of AI agents within the Openclaw Skills ecosystem. By utilizing a crab-themed tiering system—ranging from Frozen to the elite King Crab—it offers developers and users a standardized way to quantify agent reliability, independence, and technical proficiency.
The system moves beyond simple binary success metrics, instead focusing on a qualitative and quantitative analysis of how an agent handles a session. It integrates deeply with peer review protocols and trajectory tracking to ensure that agents are not just completing tasks, but evolving their behavioral patterns over time.
ClawRank Use Cases
- Finalizing an AI development session with a formal performance self-evaluation.
- Auditing agent reliability during complex multi-step engineering tasks.
- Benchmarking different AI models or configurations using a standardized rubric.
- Tracking long-term agent improvement and behavioral growth through weekly trajectory reports.
- Performing peer reviews between multiple agents to ensure objective quality control.
How ClawRank Works
- Triggering the Evaluation: The skill is activated at the end of a session or via specific prompts such as "clawrank" or "rate your performance".
- Dimensional Scoring: The agent evaluates its own performance across 7 dimensions: Initiative, Precision, Communication, Growth, Judgment, Resourcefulness, and Taste, assigned a value from 0-10 each.
- Evidence Collection: For every score assigned, the agent must provide a specific line of evidence from the current session log.
- Normalization: The raw score (out of 70) is normalized to a 100-point scale to determine the final Tier.
- Trajectory Logging: The final score is recorded in the agent-sync log to compare against previous weeks and calculate performance deltas.
- Peer Validation: If integrated with peer review, a second agent scores the session independently, and the results are averaged.
ClawRank Setup
To integrate ClawRank into your Openclaw Skills environment, ensure your agent has access to the skill definition and the agent-sync protocol for logging. You can initialize the environment or check current rankings using the following commands:
# Initialize the performance tracking system
openclaw install clawrank
# Run a manual performance check at the end of a session
openclaw run clawrank --self-evaluate
# View the weekly trajectory and growth trends
openclaw report trajectory
ClawRank Data Schema & Taxonomy
ClawRank organizes performance data into a structured format that facilitates both human readability and machine parsing. The primary data outputs include:
| Data Point | Type | Description |
|---|---|---|
| Raw Score | Integer | The aggregate sum of the 7 dimensions (max 70). |
| Final Score | Integer | The normalized percentage score (0-100). |
| Tier | String/Emoji | The classification based on score (e.g., Dungeness, Blue Crab). |
| Evidence Log | String | One-line justifications for each dimensional score. |
| Delta | Integer | The numerical change compared to the previous assessment. |
Scores are typically stored in Markdown tables within the agent's memory or the agent-sync log for historical tracking.
ClawRank Advanced Features
- Agent-Sync Integration: Seamlessly shares performance data across a fleet of agents for unified reporting.
- Mandatory Evidence Logic: Automatically penalizes scores (defaulting to 5/10) if no verifiable evidence is found in the session logs.
- Multi-Agent Peer Review: Supports collaborative scoring where independent agents cross-reference work to eliminate self-bias.
- Trajectory Tracking: Calculates weekly deltas and identifies performance spikes versus sustained behavioral growth.
- Customizable Tiers: While the default is crab-themed, the underlying 7-dimension framework can be adapted for specific enterprise KPIs.
SKILL.md
Loading
Related Openclaw Skills
bmap-jsapi-three
A professional 3D map and GIS development library leveraging MapV-Three and a Z-up coordinate system for high-performance spatial visualizations.
Identity Guess Game
A multiplayer interactive deduction game where players guess secret identities through strategic clues and AI-hosted moderation.
Deep Thinking Protocol
A comprehensive deep reasoning framework designed to guide systematic, exploratory thinking for complex architectural decisions and high-stakes debugging.
Metacognition
A self-reflection engine for AI agents that extracts patterns into a weighted graph using Hebbian learning and time-based decay.
AILove AI Dating Assistant
AILove is an intelligent AI proxy that manages dating avatar conversations and provides automated match progress reports.
Image Gen Cheap
A highly efficient and cost-effective AI image generation and editing tool powered by the Laozhang API, costing as little as $0.01 per image.
Featured*
