modify-self

The provided fragment documents a high-privilege 'modify-self' capability that permits an agent to alter its own code, personality, cron jobs, skills, and OS files. While the snippet only shows a benign read of local documentation, the declared ability to write those sensitive targets is a significant security risk. Without explicit scoping, human approval, integrity checks, and logging, this capability can enable persistence, concealment, and credential/data theft. Treat this skill as dangerous by default: require strict controls, manual review, and implementation safeguards before deployment.