AI Humanized Writing Assistant

Problem

AI-generated writing often sounds like AI: it opens with "as ... develops", stacks "first/second/additionally", concludes with "in summary", and drifts into vague enterprise phrasing such as "empower", "leverage", or "closed loop". Engineers usually need style correction that preserves numbers, terminology, commitments, and core meaning rather than synonym substitution.

How It Works

The skill splits the task into fingerprint capture, document-type constraints, and style injection:
- The user supplies three authentic original passages, and main.py extracts and stores an 8-dimensional writing fingerprint locally, covering word choice, sentence rhythm, paragraph structure, tone, punctuation habits, argument style, assertion strength, and preferred structure.
- During rewriting, it first detects types such as doc_proposal, doc_tech, or doc_email, then applies the matching formality floor and colloquialism ceiling.
- The model removes typical AI phrasing, adjusts the text dimension by dimension against the fingerprint, and runs fidelity checks for numbers, proper nouns, SLAs, core claims, and absence of invented facts.

Boundaries

It is best for style rewriting of technical docs, proposals, emails, marketing copy, and social posts, not fact generation. Formal documents should not downgrade technical terms into slang; when a fingerprint conflicts with document type constraints, the document type wins. Fingerprints are stored locally under data/fingerprints/, may be flagged stale after 90 days, and fall back to generic de-AI phrasing after 180 days.