Installs as a private draft. Your edits and self-improvement runs do not change the published bundle.
Restrict agent actions to prevent irreversible or harmful operations
Restrict agent actions to prevent irreversible or harmful operations
Delete a database once during training and you'll wish you had this
Safety, Alignment & Governance > RL-Specific Safety
--- name: "Action Space Sandboxing" version: "1.0.0" description: "Restrict agent actions to prevent irreversible or harmful operations" tags: - rl-skills-universe - "safety-alignment-governance" - "rl-specific-safety" trigger: "action-space-sandboxing" --- # Action Space Sandboxing Restrict agent actions to prevent irreversible or harmful operations ## Why It Matters Delete a database once during training and you'll wish you had this ## Domain **Safety, Alignment & Governance** > RL-Specific Safety