Run a collaborative session that actively works toward user–Claude alignment on a plan, design, or PR review. Bidirectional and user-paced — Claude first probes the user's foundational knowledge in chat (teaching where needed), THEN writes the alignment doc reflecting what was learned, THEN walks through each claim with the foundation already in place. Tracks state in docs/<topic>-alignment.md. Use when the user says "align me", "work towards alignment", "make sure we're on the same page about <plan>", "I need to understand this before approving", or invokes /work-towards-alignment.
