Summary

Anbeeld/WRITING.md offers a sophisticated "humanizer" ruleset designed to elevate AI-generated text beyond generic output. It targets writers and developers leveraging AI assistance, aiming to produce sharper, genre-aware prose enriched with concrete details. This approach enhances readability and mitigates common LLM tells, providing a structured method for refining AI drafts into more polished, publishable content.

How It Works

The core methodology integrates established writing standards, web readability research, and empirical analysis of LLM outputs. It meticulously calibrates voice and format to specific genres, ensuring opinion pieces sound personal while summaries remain neutral. The ruleset replaces vague statements with verifiable details, actively combats templated prose patterns like repetitive cadence or overly neat paragraph structures, and prioritizes precise language over inflated phrasing or artificial humanization tricks.

Quick Start & Requirements

Installation involves copying the /skills/writing directory to the global skills folder of compatible AI agent tools such as Claude Code, Codex, OpenCode, or Cursor. The skill is designed for on-demand loading, minimizing context window usage. The ruleset is scoped for public-facing prose, including articles, documentation, and marketing copy, explicitly excluding code comments or private notes.

Highlighted Details

• Self-auditing workflow: Promotes iterative refinement via AI-driven audits against the ruleset's criteria.
• Genre-aware calibration: Dynamically adapts voice, tone, and structure to specific contexts (e.g., opinion pieces vs. technical summaries).
• Concrete anchors: Replaces polished generality with checkable details, actively guarding against fake specificity.
• Agent integration: Packaged as a reusable skill for seamless incorporation into various AI agent harnesses and workflows.

Limitations & Caveats

While significantly improving AI text quality, the output may not consistently pass expert human review or advanced AI detection tools, reflecting the inherent limitations of prompt-level instruction. The ruleset is primarily focused on public-facing prose and does not extend to internal documentation or code-related text.