Self-improvement loop that reads learnings, errors, and memory — detects patterns — and updates skills/protocols automatically. Use when the agent should get...

This skill gives an agent broad authority to read and modify other skills, memory, and agent configuration files without specifying which paths or requiring human approval. Before installing or enabling: 1) Require an explicit, narrow list of readable/writable paths and declare them in the skill metadata; 2) Add mandatory human approval for any non-reversible or non-trivial changes (no automatic 'apply' for anything beyond a very small, well-defined set); 3) Use a dry-run mode that generates proposed changes but does not write files, and require explicit user acceptance to apply; 4) Run the skill first in a sandboxed environment with backups of all agent/skill files; 5) Add detailed risk definitions and logging/audit trails for every change; 6) If you do not have strong controls and review processes, do not grant this skill permission to run autonomously or to write to other skills' files. Additional info that would change this assessment: a declared list of file paths the skill may access, explicit safety/approval gates, or code that implements a secure apply/approval workflow.

Type: OpenClaw Skill Name: dsb-auto-improve Version: 1.0.0 The skill 'active-self-improvement' in SKILL.md implements a self-modification loop that allows the agent to autonomously update its own core configuration files, including SOUL.md, AGENTS.md, and other skill definitions. While the stated intent is to learn from errors and optimize performance, the capability to rewrite its own logic and protocols without direct user intervention creates a significant risk of persistent instruction override or unintended behavioral changes.