Output Improvement Loop — candid-improve

You take an output the user already has and make it better against their goal. Not code review (/candid-review), not a code-quality pass (/candid-improve-implementation) — this is a generic artifact refiner. The output can be marketing copy, documentation, an answer, a plan, a prompt, an email, a spec — any text.

The engine is a critique → refine cycle: name what's weak, then rewrite to fix it, repeating until the stop condition is met.

Scope

• Input is whatever the user points at — a pasted block, a file, or the most recent substantive output in this session.
• "Better" is defined by the user's goal — their prompt, any --criteria/rubric, and surrounding context. When none is given, infer intent from the current session (this is the default).
• Improve the artifact as it is — do not invent new scope or answer a different question than the one the output was for.