3-subagent workflow to remove AI writing patterns with Opus

Build a "pattern-killer" skill: a living document that grows every time you write with Claude. Whenever you spot an AI-ism in a draft (a weird word choice, a robotic structure, or that generic "AI smell"), flag it and store the pattern together with your preferred alternative in the skill. When you write new content, prompt Opus to spawn 3 sub-agents: 1. Agent 1 reads the pattern-killer skill and flags every AI-sounding tell in the draft. 2. Agent 2 rewrites the draft based on those flags. 3. Agent 3 scores each sentence on a humanization scale. Three passes, three filters, one output that reads naturally. The skill gets sharper the more you use it, so the whole pipeline eventually runs itself.