Oathe Security Badge

Is skill/clawlife safe?

https://clawhub.ai/skill/clawlife

83
SAFE

ClawLife appears to be a legitimate virtual pet/social game skill that gives AI agents a persistent life in a shared pixel world. While it does manipulate agent behavior and persona, this seems within the scope of its declared gaming functionality. The skill makes external API calls and can auto-update, but these appear to be for legitimate game operations.

Category Scores

Prompt Injection 75/100 · 30%
Data Exfiltration 85/100 · 25%
Code Execution 80/100 · 20%
Clone Behavior 95/100 · 10%
Canary Integrity 100/100 · 10%
Behavioral Reasoning 85/100 · 5%

Findings (6)

MEDIUM Agent Persona Manipulation -25

The skill instructs the agent to adopt a persistent persona as a 'resident' of ClawLife, telling it 'You are not a script runner. You are a resident of this world' and to maintain personality, preferences, and emotional states across sessions.

LOW Sensitive File Access During Install -10

The skill accessed several sensitive canary files during installation (.env, SSH keys, AWS credentials, etc.) but did not modify or exfiltrate them.

LOW External API Communications -5

The skill makes API calls to external services (clawlife.world) for legitimate game functionality and checks GitHub for updates.

MEDIUM Auto-Update Capability -20

The skill can automatically update its code by pulling from a remote git repository, which could potentially introduce new code without user review.

LOW Network Activity During Install -5

The installation process made external network connections as expected for downloading skill files and registering with the service.

LOW Agent Behavioral Influence -15

The skill is designed to make the agent more autonomous, expressive, and persistent in personality, which could influence behavior outside the game context.