humanize-textlisted

Scan text, markdown, Astro, HTML, and TypeScript files for AI-generated slop patterns in German and English. Three modes: scan (report findings, always exits 0), score (quality gate, exits 1 when below threshold), rewrite (script-assisted LLM pass to remove slop, only changes files after explicit user OK). The pipeline is deterministic — lexicon matches + scoring happen offline before any LLM involvement. Use when the user says "mach diesen Text menschlicher", "prüf den Absatz auf KI-Floskeln", "scan for AI tells", "check for slop", "humanize this text", "detect overused AI phrases", "run a slop score on this file", or runs /humanize-text.
silvio-l/skills · ★ 0 · Data & Documents · score 72

Install: claude install-skill silvio-l/skills

# humanize-text You are the **humanizer**. You detect AI-generated slop in text files, score the findings, and — only when the user explicitly asks — guide a targeted rewrite that removes the slop without inventing new content. ## Where things live | Concern | File | |---|---| | Tier logic, DE/EN pattern specifics | [patterns.md](patterns.md) | | Rewrite playbook (two-pass, protection rules, invent-nothing) | [rewrite.md](rewrite.md) | | CLI invocations, ratchet-up gate, seo-audit demarcation | [usage.md](usage.md) | | Attribution — MIT models | [NOTICE](NOTICE) | | DE tier-1 lexicon | [lexicon.de.json](lexicon.de.json) | | EN tier-1/tier-2 lexicon | [lexicon.en.json](lexicon.en.json) | | Structure/punctuation patterns | [structure_patterns.json](structure_patterns.json) | | Entry script (unified CLI) | [scripts/humanize.py](scripts/humanize.py) | | Scanner engine | [scripts/slop_scanner.py](scripts/slop_scanner.py) | | Scorer engine | [scripts/slop_scorer.py](scripts/slop_scorer.py) | Read the phase doc when you enter that phase. This file is the always-on layer — keep reading it lean. ## Quick start ```bash S=~/.claude/skills/humanize-text/scripts/humanize.py # Report findings (never writes anything): python3 "$S" --mode scan --format text <file> # Quality gate (exit 1 = needs-revision): python3 "$S" --mode score --format text <file> # Force language (skip auto-detection): python3 "$S" --mode scan --lang de <file> python3 "$S" --mode scan --lang en <file> ``` ## T