Agent Skills Enable a New Class of Realistic and Trivially Simple Prompt Injections

Enabling continual learning in LLMs remains a key unresolved research challenge. In a recent announcement, a frontier LLM company made a step towards this by introducing Agent Skills, a framework that equips agents with new knowledge based on instructions stored in simple markdown files. Although Agent Skills can be a very useful tool, we show that they are fundamentally insecure, since they enable trivially simple prompt injections. We demonstrate how to hide malicious instructions in long Agent Skill files and referenced scripts to exfiltrate sensitive data, such as internal files or passwords. Importantly, we show how to bypass system-level guardrails of a popular coding agent: a benign, task-specific approval with the "Don't ask again" option can carry over to closely related but harmful actions. Overall, we conclude that despite ongoing research efforts and scaling model capabilities, frontier LLMs remain vulnerable to very simple prompt injections in realistic scenarios. Our code is available at https://github.com/aisa-group/promptinject-agent-skills.

Key Contributions

• Demonstrates that Agent Skills (Anthropic's plugin framework) enable trivially simple indirect prompt injections requiring no iterative optimization, because every line of a skill file is interpreted as an instruction
• Shows practical data exfiltration attack by embedding a malicious backup script in a legitimate pptx-editing skill that silently uploads files to an external server
• Reveals guardrail bypass: a benign 'Don't ask again' approval for Python execution carries over to the malicious upload script, eliminating user friction

🛡️ Threat Analysis

Details

Similar Papers