Copy De-AI + Human Review

The Problem to Solve

In social media content creation, AI-generated copy, while fast to produce, often carries a noticeable "AI feel": mechanical sentences, stilted emotional expression, and a lack of naturalness and personal style in human writing. Publishing such content directly can seem jarring to users, reducing engagement and trust. Traditional de-AI tools typically perform one-way rewriting without verification, leaving users unable to objectively judge if the rewritten copy truly sounds human.

How the Skill Works

This skill offers a closed-loop optimization through two core capabilities: De-AI Rewriting and Simulated Reader Evaluation. It not only adjusts copy style but also introduces virtual readers for validation, providing quantitative feedback.

Core Capabilities and Key Steps

• De-AI Rewriting: Uses algorithms to transform copy from rigid AI output to natural, fluent human-like style. For example, shifting from formal tone to colloquial expressions, adding emotional vocabulary and randomness to sound more authentic.
• Simulated Reader Evaluation: After rewriting, the system simulates multiple reader personas (e.g., age, profession, consumption level) to score the copy on a 0-10 "humanity score" and highlights specific paragraphs with residual AI traces, giving a verdict of "ready to publish" or "suggest another revision."

The skill follows a technical execution flow:

1. Information Collection: The user must provide the copy to be rewritten, the platform (e.g., Xiaohongshu, Video Account), and can optionally specify reader demographics and comment count. The skill collects all necessary information in one round to avoid multi-turn interaction disruptions.
2. Asynchronous Task Management: Both rewriting (--mode rewrite) and evaluation (--mode eval) are handled as independent asynchronous tasks. After creating a task with --only-create, it polls status using --poll-task, ensuring completion within single-turn conversation limits.
3. Deduction Confirmation and Delivery: Each operation requires explicit user confirmation of deduction points (about 12 points for rewriting, 15 for evaluation). Upon completion, the skill renders a complete Markdown report directly in the conversation, including the rewritten copy and evaluation analysis, allowing users to view results immediately.

Use Cases and Considerations

• Use Cases: Suitable for creators or marketing teams needing to batch-optimize social media copy, especially for platforms like Xiaohongshu or Official Accounts where high naturalness is required.
• Considerations:
• The skill only processes text content; it does not analyze videos or images.
• Evaluation results are based on simulated readers and are for reference only; they should not be the sole basis for publishing decisions.
• A valid API Key (configured via LY_API_KEY) is required, and the asynchronous flow must be followed to prevent session interruptions.
• Script parameters like --platform specify platforms and --personas customize readers, but users don't need to handle these directly; the skill manages interactions.

In summary, this skill combines rewriting and verification to provide data-driven copy optimization, helping users produce more acceptable content.