Self-Improving Global Safe

• Risk level: Low
• Analyzed: Mar 3, 2026, 1:01 AM
• Findings: 0
• Version: 0.1.0

Overview

The skill's declared purpose (a local, namespaced learning memory) matches its instructions and requirements: it only reads/writes under a defined home-folder root, requires no external credentials, and has explicit privacy guardrails.

This skill appears coherent and implements a local memory stored under ~/.openclaw/self-improving/, with explicit rules for learning, promotion, and deletion and with stated privacy guardrails. Before installing or enabling it: (1) confirm you are comfortable with the agent reading workspace and Git metadata and creating files under your home directory; (2) test in a limited project first and inspect the created files to verify behavior; (3) be cautious about accidental storage of secrets — the docs say 'never store credentials', but that is an instruction, not an enforced technical control; (4) if you need stronger guarantees, restrict the agent's filesystem permissions for the runtime (or require encryption of the memory files) and require explicit user confirmation before any global promotion or cross-context read; (5) ask the agent to show the exact paths and a sample of the files it will create before giving it write permission.