Write, refactor, and evaluate assistant prompts, tool-use instructions, and agent workflows; diagnose LLM behavior regressions.
