Humanizer Skill: De-AI Your Agent’s Professional Writing

Where this fits

You just finished a draft for a help centre article, but the response reads like it was assembled by a generic LLM: “In today’s fast-paced world”, “delve into”, and an em dash after every second thought. You need the information to stay professional and accurate, but the tell-tale AI rhythm has to go. Instead of sending the agent back with vague instructions, you can invoke a skill that knows the exact patterns to remove.

Why agents benefit

• Detects concrete AI tells—inflated symbolism, promotional language, -ing analyses, vague attributions, em dash overuse, and rule-of-three lists—instead of doing a generic “make this sound human” rewrite.
• Preserves professional register; output stays formal, specific, and credible rather than becoming casual or conversational.
• Fixes surface patterns without restructuring meaning, so technical details and compliance language remain intact.
• Can be used as an explicit review step in an agent loop, letting the agent self-edit before a human sees the text.
• Triggers on natural commands like humanize, de-AI this, or fix the AI tone, making it drop into existing agent workflows.

Practical scenarios

Forum thread responses at support time

A support agent drafts replies to customer threads. Before posting, it runs humanizer to remove rule-of-three constructions and promotional phrases, leaving an answer that sounds like a senior engineer wrote it.

Help centre articles before publication

Your editorial agent generates a first draft of a troubleshooting guide. Running humanizer removes the “not just X, but Y” negative parallelisms and AI vocabulary, so the article reads like standard technical documentation without a corporate spin.

API documentation cleanup

You inherit documentation containing phrases like “unlock the power of” and “empower developers”. The humanizer skill edits those strings while preserving endpoint descriptions and command names, making the docs feel neutral and maintainable.

Add it to your agent workflow

Install the skill with:

npx skills add org-quicko/skills --skill humanizer

Then trigger it in a review step:

user: humanize this draft before we publish:
Our dashboard empowers teams to seamlessly delve into their metrics.
assistant: Our dashboard gives teams a direct view of their metrics.

Set it as a separate review pass rather than a global rewrite so it only fires when the user asks for de-AI editing.

Compare before adopting

• Quality signals: Run it on a sample full of em dashes, “delve”, and rule-of-three phrases. Check whether edits preserve meaning, remove the tells, and still sound professional.
• Maintenance freshness: Is the pattern list kept current? A static skill will miss new AI vocabulary as local models and prompting styles evolve.
• Alternatives: Compare against a zero-shot prompt like “rewrite this naturally”. A dedicated skill should be more consistent and less likely to over-casualize the text.
• Workflow fit: Does it integrate with your agent’s current editor flow, or does it require model-specific tooling? Check repo dependencies and licensing.

Why it is worth tracking

This skill is early—it has not accumulated community stars yet—but its practical value is in the concrete editing rules. If you routinely publish or review agent-generated technical and professional text, watch whether it removes specific AI patterns without flattening the author’s voice. Track commit activity and issue responses as signals of maintenance quality. The full skill card is available on OpenAgentSkill: https://www.openagentskill.com/skills/org-quicko-humanizer. Re-evaluate after a few real editing passes: if it consistently catches patterns that generic prompts miss, it earns a permanent place in your agent workflow.