ai-writing-detector

Installable writing skills for AI agents that need to sound like a careful writer, explain a procedure plainly, or check a draft for familiar AI prose signals. Choose the skill for the job in front of you; a mixed document does not need to pass through the whole suite.

Current source: 2.12.16

The humanizer catalog now protects the writer's judgments. A stated reaction, preference, or ranking counts as source meaning even when the reason is missing, a repair keeps the scope the source claimed, and an empty concession or contrast is reported instead of invented.

The detector keeps house-style typography out of authorship scoring. Curly punctuation, a prose em dash, and an Oxford comma together report at low severity as a style check in plain text, and they cannot move the score, label, or confidence. The "X is dead; long live Y" phrase is editorial judgment now, so essay titles and supported contrasts stay clean.

The latest GitHub release is 2.12.16; the current source tree matches it.

See CHANGELOG.md for the full history.

Choose a Skill

Use one skill for the job a section needs.

What You Can Do

• Rewrite without replacing the writer. Humanizer keeps facts, technical terms, uncertainty, point of view, and deliberate style while cutting filler.
• Control technical form when clarity is the priority. Simple English offers strict and STE-flavored modes plus a Python linter.
• Measure signals without making authorship claims. AI Writing Detector reports issue bands and can compare a before/after pair for damaged code, URLs, or structure.
• Pair editorial judgment with a repeatable gate. Writing Prose ships its Vale configuration, house rules, and long-form references with the skill.
• Match the audience. Writing Voice picks Chat, Human, or Worker, then optional tone overlays. It routes to Humanizer's voice-profile table and does not invent a personality the source never had.

Minimal Examples

Humanizer is a writing workflow rather than a command-line filter. Install the skill, then ask the agent to rewrite, clean up, critique, detect, or minimally edit supplied text. Improve the writing without inventing facts or a new personality.

Simple English includes a deterministic linter:

python3 skills/simple-english/scripts/voice_lint.py path/to/draft.md

AI Writing Detector exposes a zero-dependency Node.js scan and a preservation check:

node skills/ai-writing-detector/scripts/analyze.js path/to/file.md
node skills/ai-writing-detector/scripts/validate-cli.js before.md after.md
bash skills/ai-writing-detector/scripts/smoke.sh

Writing Prose runs the bundled Vale rules through one wrapper:

bash skills/writing-prose/scripts/vale-lint.sh path/to/draft.md

Detector output is signals_only. It is not proof of authorship and should not be used as a score to chase.

Install

Each directory under skills/ is self-contained. Copy only the skill you need:

cp -R skills/humanizer ~/.claude/skills/

Other supported skill directories include ~/.codex/skills/ and ~/.openclaw/skills/.

For a Hermes category layout:

mkdir -p ~/.hermes/skills/writing
cp -R skills/humanizer ~/.hermes/skills/writing/
cp -R skills/simple-english ~/.hermes/skills/writing/
cp -R skills/ai-writing-detector ~/.hermes/skills/writing/
cp -R skills/writing-prose ~/.hermes/skills/writing/
cp -R skills/writing-voice ~/.hermes/skills/writing/

Or install an individual package with skills.sh:

npx skills add apoapostolov/humanizer --skill humanizer
npx skills add apoapostolov/humanizer --skill simple-english
npx skills add apoapostolov/humanizer --skill ai-writing-detector
npx skills add apoapostolov/humanizer --skill writing-prose
npx skills add apoapostolov/humanizer --skill writing-voice

Repository Layout

skills/
├── humanizer/
├── simple-english/
├── ai-writing-detector/
├── writing-prose/
└── writing-voice/

Each package contains its own SKILL.md, agent metadata, references, and any runtime scripts it needs. Maintainer guidance is in AGENTS.md.

Support

Support, feedback, and feature ideas: @ApoMakesMods on X.

License

This repository is licensed under the MIT License.

Simple English source limits and credits are documented in source-and-limits.md. Detector engine attribution is in ATTRIBUTION.md.