IMPERIAL Text Hygiene + Natural Editorial Skill

Purpose: clean user-owned text of suspicious/invisible Unicode and machine-like formatting artifacts, then apply natural professional editorial quality without changing facts.

Deterministic hygiene:

• normalize line endings and Unicode where safe;
• remove zero-width space, BOM/word-joiner, soft hyphen and bidi override/isolate controls that can create hidden text behavior;
• preserve ZWJ/ZWNJ because they are meaningful in emoji and several writing systems;
• report removed-codepoint counts and never claim provenance from a cleanup result.

Editorial pass:

• remove formulaic filler, repeated structures and awkward machine cadence;
• vary sentence length naturally;
• prefer concrete wording over generic grand claims;
• preserve facts, names, numbers, technical identifiers, approvals and legal meaning;
• Privacy Shield runs before any public output;
• any change after Architect approval invalidates that approval.

Allowed intent: readability, privacy hygiene, removal of hidden characters/metadata from content the user controls, brand voice and professional editing.

Not a promise: do not claim the output is human-authored, undetectable, or guaranteed to defeat an AI detector. Do not use this skill to fabricate authorship or bypass required provenance/compliance disclosures.

Truth boundary: TEXT_CLEANED != HUMAN_AUTHORED; NATURAL_STYLE != AI_DETECTOR_EVASION; HIDDEN_CHARACTER_REMOVAL != PROVENANCE_ERASURE.

Source note: guillaumemeyer/watermarks-remover is MIT and includes an Agent Skill for inspecting/cleaning AI-related marks; IMPERIAL only adopts the safe text-hygiene/editorial subset here, not a blanket anti-detection promise.