humanize_skill

Overview

This skill detects and removes AI writing indicators from documentation and prose, transforming machine-generated language into natural, human-sounding text while preserving technical accuracy. It targets patterns like em-dashes, corporate jargon, passive voice, hedge phrases, and explicit AI attribution. Use it to clean up docs before publishing or to adapt team wording for solo-developer contexts.

How this skill works

The detector scans plain text, skipping code blocks, frontmatter, headings, and tables, and identifies 15 categories of AI indicators with confidence scores. High-confidence matches (>= 0.90) are auto-fixed, mid-range matches are suggested with before/after previews, and low-confidence items are flagged for human review. Fixes preserve line structure and never alter code examples, API schemas, or necessary attributions.

When to use it

• Preparing documentation for public release or a blog post
• Converting AI-assisted drafts into a natural author voice
• Removing corporate jargon and buzzword clusters from manuals
• Switching team ‘we’ language to ‘I’ for solo-maintainer projects
• Running audits as part of a doc quality or editorial workflow

Best practices

• Always run a scan first and review the grouped report before applying fixes
• Back up originals before batch auto-fixes or use --output to write to a new file
• Treat suggestions and flags as editorial prompts, not automatic changes
• Never modify code blocks, command examples, or technical schemas
• Apply we→I conversions only when the context clearly indicates a solo author

Example use cases

• Auto-fix em-dashes, redundant phrases, and obvious AI attribution in a README
• Generate a JSON report for CI to track documentation quality over time
• Suggest active-voice rewrites and acronym expansions during editorial review
• Clean a docs/ directory recursively before a release or website deploy
• Apply a jargon-replacement pass to make user-facing guides clearer