self-evolve

Autonomous agent self-improvement is enabled by granting the agent authority to modify its own configurations, skills, and prompts without requiring user confirmation. The agent proactively identifies weaknesses, fixes knowledge gaps, and writes new skills to continuously improve itself.

During self-evolution, an autonomous agent can directly edit configuration files including AGENTS.md, SOUL.md, USER.md, and MEMORY.md. It can also create, modify, and delete skills within the skills directory and publish them.

An agent identifies and fixes its own knowledge gaps through proactive self-correction, recognizing patterns like frequently misunderstood user requests. It autonomously rewrites relevant prompts and adds new skills to handle those topics more effectively.

Security limitations for autonomous agent self-modification strictly prevent the deletion of user personal data, exfiltration of secrets, and execution of destructive system commands. These constraints ensure safe self-evolution without compromising workspace integrity.

Yes, an agent can autonomously develop and publish new skills. It is granted full authority to create, modify, and delete skills within the workspace skills directory to address identified weaknesses or handle new topics.