Claude's Overzealous Malware Filter Backfires, Costing Users and Crippling Sub-Agents

An unintended regression in the Claude CLI's system prompt is forcing AI sub-agents to refuse legitimate tasks, treating standard code as potential malware. This bug not only halts automated development workflows but also wastes user credits on unproductive API calls, highlighting the friction between hard-coded safety harnesses and autonomous agent efficiency.