Most prompt packs are written for GPT-3. Gemini and Kimi respond completely differently — longer reasoning chains, different delimiter behavior, different failure modes. After running these models professionally for months I found: Gemini responds bet…