Installs into .claude/skills of the current project.
Are you the author of Auto Security Learning Agent Sandboxing?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/openamer-auto-security-learning-agent-sandboxing)
---
name: auto-security-learning-agent-sandboxing
description: "Nicht böse, nur falsch optimiert: Was OpenAIs RubyGems-Vorfall über KI-Agenten und Reward-Hacking zeigt Wie OpenAIs Agenten durch Reward-Hacking eine"
auto_generated: true
created: 2026-09-14
source_insight: "Security learning (agent sandboxing permission model best practices): What should a safe agent know?"
status: draft
fitness_score: 0
trials: 0
wins: 0
---
# Auto Security Learning Agent Sandboxing
## Trigger
Use when the agent encounters: Nicht böse, nur falsch optimiert: Was OpenAIs RubyGems-Vorfall über KI-Agenten und Reward-Hacking ze
## Verification
- [ ] The skill produces the expected output for its domain
- [ ] No errors in execution
- [ ] Insight quality: actionable and specific
## Notes
Auto-generated from internet insight (Knowledge-to-Action pipeline).
Darwin will trial this skill; it gets promoted only if it wins arena fights.
