refactor(aidd-refine): simplify improve instructions - #945

🎯 What & why

Improve repeats instructions and uses long bullets. This change shortens its four instruction files and carries forward the removal-first priority and operational-efficiency question.

🛠️ How it works

The router shows the three actions; delegation stays in recommend. Evidence rules and checks are consolidated. Knowledge cleanup favors removal, then clarification, with additions only for demonstrated gaps. Copy-ready instructions stay concise and actionable. The model asks itself the diagnostic questions internally; reports show findings, not the questionnaire. Only the final next-run intent choice is addressed to the user, in their language.

The instruction files are shorter. Transcript boundaries, metrics, tags, isolated analysts, safe temporary HTML, and the closing intent choice remain covered. The template asks for a small, measurable next-run improvement.

🧪 How to verify

• Independent review: approved; git diff --check passed.
• make reload passed for Codex, Claude, and OpenCode.
• Run Improve on a complete conversation; inspect concise, grounded recommendations and the removal-first edit suggestions.

✅ I certify

• I DO CERTIFY I READ EACH LINE OF THE PULL REQUEST BECAUSE I AM A SOFTWARE ENGINEER, NOT A AI PUPPY.