Towards a Risk Assessment of Malicious Skill Files in Coding Agents