Picture this scenario: An Anthropic Skill scanner runs a full analysis of a Skill pulled from ClawHub or skills.sh. Its markdown instructions are clean, and no prompt injection is detected. No shell ...