LifelongLazyLearner/qu-ai-wei

What it solves

It addresses the problem of "AI-flavored" Simplified Chinese text—writing that feels stiff, formulaic, or overly structured (often characterized by translationese or repetitive corporate jargon). The project aims to make such text sound more natural and human-like while strictly preserving the original facts, judgments, formality level, and the author's voice.

How it works

Instead of simple word replacement, the tool follows a structured multi-step process:

1. Validation: Checks for credentials, authorization, and intended use.
2. Locking: Protects critical data like numbers, names, dates, technical terms, and causal relationships to prevent "hallucinations" or drift.
3. Pattern Scanning: Scans for eight families of common AI writing patterns (e.g., empty introductory phrases, mechanical symmetry, and excessive connectives) and removes them.
4. Restructuring: Reorganizes the flow of facts, explanations, and conclusions, including adjusting paragraphing and headings for longer texts.
5. Verification: Performs a final review to ensure that evidence strength and original intent remain intact.

Who it’s for

Users who write or edit Simplified Chinese content and want to remove the robotic quality of AI-generated text without losing technical accuracy or professional tone.

Highlights

• Fact Preservation: Explicitly prevents the alteration of numbers, technical terms, and specific details.
• Context Awareness: Distinguishes between technical documentation (which should remain formal) and personal narratives (which should remain natural).
• Pattern-Based Approach: Targets specific linguistic markers of AI writing rather than using a generic rewrite prompt.
• Broad Tool Compatibility: Uses the Agent Skills format, making it compatible with tools like Claude Code, Cursor, Kimi Code CLI, and Codex.