A Claude Code skill that strips AI writing patterns and rewrites content in your actual voice.
Most humanizing tools sand off the obvious tells. This one goes further. It's a consolidation of every effective humanizing technique across the best available approaches, rebuilt from scratch with significant improvements. It also knows what not to touch: code blocks, prompts, and technical content are left alone. Only the human-facing copy gets rewritten.
The result reads like a person wrote it, not like a person tried to hide that a machine wrote it.
Everything in one place. Most approaches tackle one thing at a time: swap the vocabulary, vary the sentence length, remove the em dashes. This skill runs the full playbook in a single pass. Content inflation, language slop, structural tells, formatting artifacts, chatbot residue: all of it, simultaneously.
It doesn't touch your code or prompts. Code blocks, technical commands, and prompt text are detected and skipped. The skill rewrites the writing around them, not through them.
It learns your voice. Give it your podcast transcripts, book chapters, or Instagram captions once. It analyzes how you actually write: your sentence rhythm, your recurring phrases, your perspective, what you never say. Every rewrite after that sounds like you, not like a cleaned-up chatbot.
The two most common locations:
~/.claude/skills/
~/.config/claude/skills/
Not sure which one? Check your Claude Code settings, or look for an existing skills/ folder.
git clone https://github.com/josephtandle/speak-human
cp -r speak-human/speak-human ~/.claude/skills/speak-humanSymlink if you want updates to pull through automatically:
git clone https://github.com/josephtandle/speak-human
ln -s "$(pwd)/speak-human/speak-human" ~/.claude/skills/speak-humanIn a Claude Code session:
/speak-human "This comprehensive guide delves into the intricacies of our platform."
You should get a clean rewrite back.
/speak-human "your text here"
/speak-human --file draft.md
/speak-human --file post.md --mode detect
/speak-human --file email.md --mode edit
/speak-human "text" --voice blunt --aggressive
| Mode | What it does |
|---|---|
rewrite |
Full rewrite. Default. |
detect |
Reports all AI patterns found. No changes made. |
edit |
Applies changes in-place to the file. |
Five presets to match the format and audience:
| Voice | Best for |
|---|---|
casual |
Social posts, newsletters, conversational content |
professional |
LinkedIn, business writing, proposals |
technical |
Docs, READMEs, technical posts |
warm |
Emails, community posts, personal content |
blunt |
Pitches, executive summaries, direct asks |
This is the most useful feature. Run it once, and every rewrite after that sounds like it came from you specifically, not from a generic "humanized" default.
Any text file that represents your voice at its best:
- Podcast transcripts
- Book or course chapters
- Instagram captions you're proud of
- Newsletter issues
- Emails that landed well
One strong file is enough to start. More content means a more accurate profile.
/speak-human --build-voice path/to/transcript.txt path/to/chapter.md
The skill reads your files, extracts your voice patterns, and saves a permanent profile to ~/.speak-human/voice-profile.md. It asks before overwriting an existing profile.
To rebuild from scratch without the prompt:
/speak-human --rebuild-voice path/to/new-content.txt
/speak-human "your AI-generated draft" --my-voice
/speak-human --file blog-draft.md --my-voice
/speak-human --file email.md --my-voice --aggressive
Your profile replaces the preset voices entirely. The rewrite is calibrated to match your rhythm, vocabulary, and tendencies.
- Sentence length distribution and rhythm
- Paragraph structure tendencies
- Signature vocabulary and recurring phrases
- Perspective (I / you / we)
- Hedging level
- Tone markers: humor, directness, warmth, formality
- Words and constructions you conspicuously avoid
- 3-5 gold standard sentences that represent your voice at its best
| Flag | Description |
|---|---|
--mode detect|rewrite|edit |
Operation mode. Default: rewrite. |
--voice casual|professional|technical|warm|blunt |
Voice preset. |
--my-voice |
Use your saved voice profile instead of a preset. |
--build-voice file1 [file2...] |
Build voice profile from your content. Asks before overwriting. |
--rebuild-voice file1 [file2...] |
Rebuild profile, overwriting existing without asking. |
--file path/to/file |
Read input from file. Required for --mode edit. |
--aggressive |
Shorter sentences, stronger positions, zero hedging. |
The skill scans across five categories:
Content inflation: significance inflation ("pivotal moment in history"), vague attributions ("experts say"), promotional language ("world-class", "seamless"), generic conclusions ("the future looks bright")
Language slop: the full AI vocabulary blacklist (leverage, utilize, delve, foster, pivotal, and 40+ more), empty adverbs, filler phrases, throat-clearing openers, emphasis crutches ("let that sink in")
Structural problems: binary contrasts, dramatic fragmentation, rhetorical setups, false agency, passive voice, synonym cycling
Formatting tells: em dashes, bold overuse, structured list syndrome, question-format headings
Chatbot artifacts: residue phrases ("I hope this helps", "Certainly!"), knowledge-cutoff disclaimers, comprehensive overview openers
The full pattern list lives in speak-human/SKILL.md.
examples/before-after.md — sample rewrites across email, LinkedIn, technical docs, and voice profile usage.
examples/voice-profile-example.md — annotated sample of what a built voice profile looks like.
MIT