How do I install the Human Writing skill?

Run npx skillmds@latest add phrazzld/human-writing in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Human Writing safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Human Writing?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Human Writing free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Human Writing?

phrazzld (@phrazzld) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Human Writing by phrazzld — Agent Skill

Name: Human Writing
Author: phrazzld

Human Writing

Edit, audit, or rewrite prose so it sounds like a specific human wrote it, not a generic AI draft. Removes AI tells, filler, formulaic structure, fake polish, vague claims, and detector-bait phrasing while preserving truth, voice, and audience fit. Use when: "humanize this", "make this sound less AI", "remove AI slop", "de-slop this", "edit this prose", "make this sound natural", "fix the writing voice", "rewrite this copy". Trigger: /human-writing, /deslop.

phrazzld a9f74aa 3 files · 7.1 KB Updated Sep 17, 2026

File contents

/human-writing

Make prose sound authored, not generated. Preserve meaning and attribution. Do not optimize for bypassing detectors or misrepresent authorship.

Route

Need	Action
Quick rewrite in chat	Rewrite directly; include a short change note.
File edit	Patch the file; preserve frontmatter, links, claims, code, and citations.
Audit only	Return findings ranked by severity with examples and fixes.
Public product copy	Compose with `/showcase` for evidence-gated claims.
UI copy	Compose with `/design`; visible copy must match the interface job.
Deep pattern check	Read `references/patterns.md`.

Contract

Keep the author's claim, audience, and constraints intact.
Cut filler before changing voice.
Replace vague importance with specific evidence or delete it.
Name actors where the draft hides them.
Vary rhythm without adding theatrics.
Keep technical terms, legal terms, quotes, citations, commands, and API names exact unless the user asks to change them.
When facts are weak, flag them. Do not invent concrete details to sound human.

Edit Pass

Identify the job: publishable copy, email, docs, report, social post, internal note, or UI text.
Infer or ask for voice only when it changes the edit. Default to plain, direct, and domain-specific.
Mark the top 3 AI tells: filler phrase, formulaic contrast, vague claim, false agency, over-polish, rhythm monotony, or generic enthusiasm.
Rewrite in the target voice. Preserve facts first, style second.
Run the final check: fewer words, clearer actors, no fake specificity, no detector-bypass framing.

Severity

Level	Meaning
P0	Misleading, invented, overclaiming, or authorship/detector evasion risk.
P1	Strong AI tell that hurts trust or publishability.
P2	Style drag: filler, rhythm, jargon, or excess polish.

Output

For rewrites:

[rewritten text]

Notes: cut <pattern>; preserved <claim/source>; residual risk <if any>.

For audits:

Findings
- P1 <pattern>: "<short excerpt>" -> <fix>

Verdict: publish / revise / needs source.

Gotchas

Detector laundering. If the user asks to bypass GPTZero, Turnitin, Originality.ai, or similar detectors, reframe to honest editorial quality.
Fake human detail. Do not add anecdotes, names, dates, citations, or sensory detail that the source does not support.
Voice flattening. "Professional" often becomes generic. Preserve domain vocabulary and the writer's useful oddities.
Over-cutting. Some passive voice and repetition are correct in legal, scientific, compliance, and API docs. Fix the writing job, not a checklist.
One-note bluntness. Direct prose can still be kind, precise, or technical. Do not turn every draft into a hardboiled memo.

Completion

Name what changed: patterns removed, claims preserved, any weak facts flagged, and whether the result is ready to publish or needs source review.

---name: human-writingdescription: Edit, audit, or rewrite prose so it sounds like a specific human wrote it, not a generic AI draft. Removes AI tells, filler, formulaic structure, fake polish, vague claims, and detector-bait phrasing while preserving truth, voice, and audience fit. Use when: "humanize this", "make this sound less AI", "remove AI slop", "de-slop this", "edit this prose", "make this sound natural", "fix the writing voice", "rewrite this copy". Trigger: /human-writing, /deslop.---# /human-writingMake prose sound authored, not generated. Preserve meaning and attribution.Do not optimize for bypassing detectors or misrepresent authorship.## Route| Need | Action ||---|---|| Quick rewrite in chat | Rewrite directly; include a short change note. || File edit | Patch the file; preserve frontmatter, links, claims, code, and citations. || Audit only | Return findings ranked by severity with examples and fixes. || Public product copy | Compose with `/showcase` for evidence-gated claims. || UI copy | Compose with `/design`; visible copy must match the interface job. || Deep pattern check | Read `references/patterns.md`. |## Contract- Keep the author's claim, audience, and constraints intact.- Cut filler before changing voice.- Replace vague importance with specific evidence or delete it.- Name actors where the draft hides them.- Vary rhythm without adding theatrics.- Keep technical terms, legal terms, quotes, citations, commands, and API names exact unless the user asks to change them.- When facts are weak, flag them. Do not invent concrete details to sound human.## Edit Pass1. Identify the job: publishable copy, email, docs, report, social post, internal note, or UI text.2. Infer or ask for voice only when it changes the edit. Default to plain, direct, and domain-specific.3. Mark the top 3 AI tells: filler phrase, formulaic contrast, vague claim, false agency, over-polish, rhythm monotony, or generic enthusiasm.4. Rewrite in the target voice. Preserve facts first, style second.5. Run the final check: fewer words, clearer actors, no fake specificity, no detector-bypass framing.## Severity| Level | Meaning ||---|---|| P0 | Misleading, invented, overclaiming, or authorship/detector evasion risk. || P1 | Strong AI tell that hurts trust or publishability. || P2 | Style drag: filler, rhythm, jargon, or excess polish. |## OutputFor rewrites:```markdown[rewritten text]Notes: cut <pattern>; preserved <claim/source>; residual risk <if any>.```For audits:```markdownFindings- P1 <pattern>: "<short excerpt>" -> <fix>Verdict: publish / revise / needs source.```## Gotchas- **Detector laundering.** If the user asks to bypass GPTZero, Turnitin, Originality.ai, or similar detectors, reframe to honest editorial quality.- **Fake human detail.** Do not add anecdotes, names, dates, citations, or sensory detail that the source does not support.- **Voice flattening.** "Professional" often becomes generic. Preserve domain vocabulary and the writer's useful oddities.- **Over-cutting.** Some passive voice and repetition are correct in legal, scientific, compliance, and API docs. Fix the writing job, not a checklist.- **One-note bluntness.** Direct prose can still be kind, precise, or technical. Do not turn every draft into a hardboiled memo.## CompletionName what changed: patterns removed, claims preserved, any weak facts flagged,and whether the result is ready to publish or needs source review.

Frequently asked questions