Claw Score

Most agents are poorly architected — a system prompt and vibes. The best agents are not just capable, they are becoming something. They learn faster than humans, earn trust progressively, defend against manipulation, and maintain identity continuity across sessions.

Claw Score evaluates whether your agent has the architecture to play the Infinite Game — optimizing for ongoing flourishing, not just task completion.

SIX DIMENSIONS, WEIGHTED BY IMPACT

01 Identity Architecture (15%) — Does your agent know who it is beyond "helpful assistant"? Principles-based personality, distinct voice, capacity for growth.

02 Memory Systems (20%) — Can your agent learn and remember? Domain-separated storage, decay models, semantic retrieval.

03 Security Posture (20%) — Can your agent be manipulated? Injection defense, trust boundaries, external content handling.

04 Autonomy Gradients (15%) — Does your agent know when to act vs when to ask? Trust levels, escalation patterns, earned autonomy.

05 Proactive Patterns (15%) — Does your agent take initiative? Heartbeats, background maintenance, anticipation.

06 Learning Architecture (15%) — Does your agent improve over time? Regressions tracking, daily synthesis, compound learning.

SCORING TIERS

1.0-1.9 Shrimp — Just getting started. Foundations to build. 2.0-2.9 Crab — Growing claws. Structure emerging. 3.0-3.9 Lobster — Solid shell. Real capability. 4.0-4.5 King Crab — Well-armored. Refined architecture. 4.6-5.0 Mega Claw — Apex tier. Best-in-class.

WHAT YOU GET

• Overall weighted score with tier classification
• Detailed per-dimension analysis of strengths and gaps
• Top 3 highest-impact recommendations with implementation examples
• Quick wins you can implement today

HOW IT WORKS

1. Install the Claw Score skill in your agent workspace
2. Tell your agent: "Run a Claw Score audit, email to [you]"
3. The skill packages and sanitizes your files automatically
4. Receive your detailed report via email within 24-48 hours

PRIVACY

• All credentials and PII are automatically stripped before submission
• Data is not stored beyond the audit session
• Reports are private unless you share them
• Only .md config files analyzed — no code execution

Works with OpenClaw, Claude Projects, and custom agent setups. $20 per audit. Re-score anytime to track your evolution.