autowriter — Automated Writing System

autoresearch's core is an agent loop: modify code → run → evaluate → keep/discard → loop.

autowriter maps this paradigm to writing, embedding "de-AI" into the loop itself — not post-processing after writing, but writing, purifying, evaluating, rewriting in every single iteration.

Design Philosophy

Three core principles from autoresearch, mapped to writing:

autoresearch Principle	Writing Mapping	Mechanism
Automated loop	write → humanize → evaluate → rewrite loop	Agent Loop
Quantified evaluation	6-dimension scoring function (with "human feel" dimension)	Phase 2
Failure transparency	draft log records every discarded version	Draft Log

Plus humanizer's core insight: De-AI is not post-processing polish, it's part of writing quality. The evaluation function detects AI patterns directly — rewrite if not passing, rewrite again, until clean.