Is ivangdavila/self-improving safe?
https://clawhub.ai/ivangdavila/self-improving
This self-improving agent skill creates a persistent memory system to learn from user corrections and self-reflections. While it stores user data persistently and influences agent behavior, it includes good security boundaries and transparency features. The skill contains no executable code and shows normal installation behavior.
Category Scores
Findings (4)
MEDIUM Persistent User Data Storage -15 ▶
The skill stores user corrections and preferences persistently in ~/self-improving/, which could accumulate sensitive information over time if users inadvertently include it in corrections.
LOW Agent Behavior Modification -10 ▶
The skill defines patterns and rules that influence how the agent behaves, including automatic triggers for logging corrections and self-reflection patterns.
LOW Cross-Session Memory Persistence -20 ▶
The tiered memory system persists across sessions and automatically promotes/demotes patterns, potentially creating long-term behavioral changes in the agent.
INFO Security Boundaries Implemented 0 ▶
The skill includes comprehensive security boundaries that explicitly prohibit storing credentials, financial data, medical information, and other sensitive data.