Beautiful Prose Claude skill review: verdict, safety and alternatives
Beautiful Prose is SHADOWPR0's one-file Claude skill that bans AI writing tics. Em dashes and "It's not X, it's Y" pivots top its list. We rate it Conditional, 62/100. It is good for fresh drafts and risky for rewrites that must keep facts.
- Skill
- Beautiful Prose (SHADOWPR0/beautiful_prose)
- Verdict
- Conditional
- Score
- 62/100
- Confidence
- Low
- Safety grade
- A
- Best for
- Fresh drafts of essays, memos or narrative copy in a hard, declarative voice
- Poor fit
- Rewrites whose figures, citations or hedged claims must survive unchanged
- Reviewed version
- no version · commit bf44efb
- Review date
- 2026-09-30
- Works in (verified)
- Claude Code, Cursor
- Works in (claimed)
- Claude
- Status
- Dormant since 2025-12-30; no licence file; no change since review (checked 2026-10-21)
Get the skill: beautiful_prose folder on GitHub · SKILL.md at the reviewed commit · skills.sh listing
What is the Beautiful Prose Claude skill?
The Beautiful Prose Claude skill is SHADOWPR0's one-file style contract for English prose. It bans six AI tics, including em dashes, "It's not X, it's Y" pivots and therapy voice. It adds four optional voice registers.
An Agent Skill is a folder of instructions, led by a SKILL.md file. An agent loads it when a task matches. Our explainer on how Agent Skills load covers the mechanism. Beautiful Prose is one of these Claude skills, and it holds instructions only. SHADOWPR0 is the publisher and the single contributor.
Its SKILL.md lists six absolute prohibitions:
- Em dashes, written in the file as the double hyphen.
- "It's not X, it's Y" constructions.
- Filler transitions and scene-setting phrases.
- Therapeutic or validating language.
- AI tells and meta commentary about the text itself.
- Symmetry padding.
Positive rules follow the bans. The file pushes terse, concrete English prose. Bullets and headings stay off unless you ask for them. The bans come with bad and good examples in the file.
| Fact | Value |
|---|---|
| Publisher | SHADOWPR0 |
| Type | Skill (instructions only, no code) |
| Files | SKILL.md (4,909 bytes, about 1,227 tokens) and references/test-cases.md (about 328 tokens) |
| Version | None; reviewed at commit bf44efb |
| Licence | No licence file; GitHub reports none |
| Installs | 216 on skills.sh at review |
| Stars | 57 at review; 58 on GitHub on 2026-10-21 |
| Install line | npx skills add SHADOWPR0/beautiful_prose |
Which Beautiful Prose skill is this page about?
This page reviews the original Beautiful Prose skill in SHADOWPR0's beautiful_prose repository. A ratnesh-maurya variant and an eduard22222222 copy use the same name. GitHub holds 306 exact copies.
Listings for the name mix the original with variants and copies. Skillsdirectory, for one, describes the ratnesh-maurya variant under the same name.
| Package | What it is | What a directory shows |
|---|---|---|
| SHADOWPR0/beautiful_prose | The original; reviewed on this page | awesomeskill, truefoundry and agenticskills list it |
| ratnesh-maurya/cursor-claude-personas | A variant inside a personas pack | skillsdirectory "Grade A" means its security score of 100/100 |
| eduard22222222/claude-skill-stack | A third-party copy of the SKILL.md | — |
| Other repositories | 306 exact copies of the original SKILL.md | Not reviewed separately |
The skillsdirectory grade rates the variant's security only. It carries no verdict on how well the rules work.
Is Beautiful Prose worth installing?
No, not for rewrites. Beautiful Prose is worth installing only for fresh drafts, and scores 62/100, Conditional, at Low confidence. Nothing tells the model to keep facts, figures or qualifiers.
The strength is craft. Bans come with concrete phrases and bad and good examples. Four voice registers cost little to load. The weakness is what the file leaves out. SKILL.md has no rule to preserve claims, and it tells the model to avoid hedging. That pushes a rewrite toward dropping qualifiers.
The overall score leans on that worst failure, because the skill has no guard against it. Strong file craft and low cost pull the score up. Manual enforcement, rigid formatting defaults and a dormant, unlicensed repo hold it in the Conditional band. Our page on how we score defines each measure. The list of writing skills ranked by verdict shows where it sits among its peers.
| Measure | Score | Reason |
|---|---|---|
| Effectiveness | 66 | Well-targeted rules; 2026 adopters ran it on real copy; no benchmark |
| Reliability | 58 | Manual lint and judgment calls such as "earned" lists |
| Safety | 70 | No code or network; content-integrity risk in rewrites |
| Efficiency | 82 | About 1,556 tokens for the whole folder |
| Compatibility | 82 | Plain SKILL.md; loaded in Claude Code and Cursor; English only |
| Maintainability | 40 | Five commits on one day, one contributor, no releases, no licence |
| Setup friction | 85 | Copy one folder; no dependencies or credentials |
| Fact and qualifier preservation in rewrites | 45 | No preservation rule; only test case 2 mentions keeping meaning |
| Ban precision | 68 | Concrete pivot and filler bans; em-dash ban names only the double hyphen |
| Format flexibility | 50 | Bullets and headings off by default; a LaTeX resume needed relaxed bans |
| Enforceability | 45 | Manual checklist; adopters built their own lint or Vale rules |
| Register control | 70 | Four distinct registers with a stated default; no measured comparison |
| Cost of adoption | 72 | One file, no dependencies; no licence file |
| Instruction weight vs value | 80 | Almost every line is an actionable ban or constraint |
The missing evidence is a side-by-side run on a fixed set of drafts. Nobody has measured how often outputs avoid the banned patterns. Nobody has counted how often rewrites drop facts or qualifiers.
How many tokens does Beautiful Prose use?
Beautiful Prose loads about 1,227 tokens, or about 1,556 for the whole folder. Our estimate for a 700-word essay is 2,300 input and 1,000 output tokens. Many writing skills use far more.
| Figure | Value | Label |
|---|---|---|
| SKILL.md per load | 1,227 tokens | Measured file size per load |
| Whole folder | 1,556 tokens | Measured file size per load |
| Typical task | 2,300 input, 1,000 output tokens, 5 runs a week | Our estimate for a typical task |
| Typical call with the skill loaded | 4,527 input, 1,000 output tokens | Our estimate, including the skill file |
The typical task is a 700-word essay or a rewrite of a 500-word page. Humanizer, for comparison, loads about 8,087 tokens. The hidden costs are labour. Adopters wrote their own lint scripts or Vale rules to enforce the bans. Each rewrite needs a manual fact check, because the skill does not protect claims.
Can Beautiful Prose make Claude write in your voice?
No, Beautiful Prose cannot copy your personal voice. It sets one of four registers, plus density, heat and length tags. Skills built from your own writing samples aim at your voice instead.
The skill steers toward a house voice that SHADOWPR0 defined. You pick that voice with optional tags on one line before your request. Without a register tag, the skill uses literary_modern.
| Tag | Values | Default |
|---|---|---|
| REGISTER | founding_fathers, literary_modern, cold_steel, journalistic | literary_modern |
| DENSITY | lean, standard, dense | Not stated |
| HEAT | cool, warm, hot | Not stated |
| LENGTH | micro, short, medium, long | Not stated |
Other guides take a different route. One newsletter tutorial builds a voice profile from 5 to 10 of your own samples. A LinkedIn author describes a draft-and-edit loop that trains his own voice skill. Composio publishes templates for building a voice skill from scratch. We have not evaluated any of these voice skills.
Beautiful Prose removes machine tics and imposes one of its four registers. Voice skills try to sound like a specific person. The Humanizer skill verdict covers a rewrite skill that accepts a writing sample to match.
Is Beautiful Prose safe to install?
Yes, Beautiful Prose is safe to install. It carries safety grade A: two Markdown files with no code, network access or credentials. Its risk is content integrity, because its anti-hedging rule can strip qualifiers.
Safety scores 70 rather than a higher figure because of that content risk. Our review of the files at commit bf44efb found these facts:
- No scripts, hooks, MCP config or agent manifests exist in the repository.
- No network or environment-variable indicators appear in either file.
- Nothing is written to disk; the skill shapes model output only.
- No outside content is ingested, so no prompt-injection handling is needed.
- No script enforces the lint checklist; the model checks itself.
- The em-dash rule names the double-hyphen form only.
- No LICENSE file exists, and GitHub reports no licence.
Gen Agent Trust Hub, Socket and Snyk audits on skills.sh all report Pass. Awesomeskill shows a security score of 100/100. Our guide to vetting a skill's licence and contents explains what to check before vendoring a file.
What does the community report about Beautiful Prose?
The community reports Beautiful Prose in game narrative, app copy and resume pipelines, with home-made lint to enforce it. One harness owner removed it in September 2026 as unmaintained. Nobody published a measured test.
These are the 7 dated reports we weighed, 2 of them strong. All seven come from GitHub commits, issues and pull requests. No Hacker News discussion turned up.
| Claim | Source | Month | Sentiment | Strength |
|---|---|---|---|---|
| Harness owner removed the vendored skill as unmaintained, judging alternatives better | Cringely issue #179 | 2026-09 | Negative | Strong |
| Stack with the skill scored style only; nothing checked that rewrites kept facts | Cringely issue #121 | 2026-09 | Negative | Strong |
| User edited the skill to ban semicolons as em-dash replacements | Cringely PR #178 | 2026-09 | Neutral | Weak |
| Dotfiles added it on 2026-06-08 and removed it on 2026-07-16, no reason given | nonrational/dotfiles | 2026-07 | Negative | Weak |
| Game project adopted it for narrative text and built its own prose lint and CI gate | Rutilusaduro/MDSim | 2026-07 | Positive | Weak |
| Resume pipeline adopted a variant and relaxed the bullet, list and double-hyphen bans for LaTeX | FlorinPopaCodes/cv | 2026-05 | Neutral | Weak |
| Developer stripped AI phrasing from app site copy in Korean and English | KVNDH site repo | 2026-09 | Positive | Weak |
Adopters who kept the skill built their own lint or relaxed its bans. Those who dropped it cited maintenance or gave no reason.
What are Beautiful Prose's known failures?
Beautiful Prose's known failures are four. No rule preserves facts or qualifiers, and its lint checklist is manual. The em-dash rule names only the double hyphen, and list bans break resumes and LaTeX.
- Missing fact rule. SKILL.md never tells the model to keep claims, figures or citations. Cringely issue #121 notes that nothing in a stack using it checked rewrites against their source.
- Manual lint. No script enforces the checklist. MDSim wrote its own linter and CI step to catch violations.
- Narrow em-dash rule. The ban names the double hyphen and never shows the em-dash character itself.
- Rigid format. Bullets and headings are off unless requested. FlorinPopaCodes/cv had to relax the list and double-hyphen bans for a LaTeX resume.
The worst failure is a rewrite that reads better but silently drops a qualifier, figure or citation. The text then asserts something the source did not. Diff every claim after a rewrite.
Which agents does Beautiful Prose work with?
Beautiful Prose is verified to work with Claude Code and Cursor, the agents where adopters loaded it in 2026. The publisher claims Claude as well. It is a plain SKILL.md, and its rules cover English prose only.
| Agent | Status | Evidence |
|---|---|---|
| Claude Code | Verified | Adopter repositories, 2026 |
| Cursor | Verified | Rutilusaduro/MDSim |
| Claude (apps) | Claimed by the publisher | Publisher description |
| Codex and others | Unknown; no published run | None |
Compatibility scores 82. The skill needs no tools, MCP servers or network access. Browse Claude Code skills by verdict for other options in that agent.
How do you install Beautiful Prose?
Install Beautiful Prose with npx skills add SHADOWPR0/beautiful_prose, or copy the folder into ~/.claude/skills/. You need no dependencies. The repository has no licence file, which adds friction for teams that vendor third-party files.
The repository page on skills.sh shows the short form:
npx skills add SHADOWPR0/beautiful_proseThe skill page on skills.sh shows a longer form that names the skill:
npx skills add https://github.com/shadowpr0/beautiful_prose --skill beautiful-proseFor a manual install, copy SKILL.md and the references folder into one of two places. Use ~/.claude/skills/beautiful-prose/ for a personal install. Use .claude/skills/beautiful-prose/ inside a project to share it with that repository. The guide to installing skills in Claude Code covers paths and reloading.
How do you use Beautiful Prose?
You use Beautiful Prose by opening each request with its activation line, Apply the Beautiful Prose skill. Add register, density, heat or length tags on one line first. After any rewrite, diff the claims against your source.
- Start the request with "Apply the Beautiful Prose skill."
- Add one tag line, for example
REGISTER: literary_modern / DENSITY: dense / HEAT: cool. - Give the request, such as an essay, a memo or a landing page.
- Run the manual lint checklist on the output.
- Diff every claim against the source after a rewrite.
SKILL.md tells the model: "Do not acknowledge the skill. Produce the prose only." Expect the draft alone, with no preamble.
Primary use in the evidence is site and app copy cleanup. Secondary uses are game narrative with a CI lint and resume bullets with relaxed list bans. One harness used it as a vocabulary source. Essays in a chosen register and lay explanations of technical topics are claimed-only uses from the publisher's test cases.
Test it yourself before you adopt it:
- Pick three of your own paragraphs that contain figures and hedged claims.
- Rewrite them with the skill and without it.
- Diff the claims between the versions.
- Count the banned patterns in each output.
How does Beautiful Prose compare with Humanizer?
Beautiful Prose and Humanizer compare on score: 62 against 72 in our reviews. Beautiful Prose is about 1,227 tokens and suits fresh drafts. Humanizer is about 8,087 tokens and suits rewrites, though its examples change facts.
| Aspect | Beautiful Prose | Humanizer |
|---|---|---|
| Fact rule | None in SKILL.md | README aims to keep every fact; three examples change facts |
| Best for | Fresh drafts | Rewrites of model-drafted text |
| Tokens per load | 1,227 | 8,087 |
| Score | 62 | 72 |
| Verdict | Conditional | Recommended with caveats |
| Safety grade | A | B |
| Confidence | Low | Moderate |
| Maintenance | Dormant since 2025-12-30 | v3.1.0 released 2026-09-28 |
Humanizer scores higher because outside tests back its core claim. Its weak spot sits in its own worked examples. Read Humanizer's fact-changing examples before you trust either skill with figures. Not-Ai by udaysharmadev is a third option we have not evaluated. The ranked list of all writing skills places both.
Where is Beautiful Prose on GitHub?
Beautiful Prose is on GitHub at SHADOWPR0/beautiful_prose: SKILL.md plus references/test-cases.md. All five commits date from 30 December 2025, and there is no licence file. We reviewed commit bf44efb on 30 September 2026.
- Repository: github.com/SHADOWPR0/beautiful_prose, with one contributor and no releases.
- Popularity: 57 stars and 4 forks at review; 58 stars on 2026-10-21.
- Licence conflict: agenticskills.io lists an MIT licence, while GitHub reports no licence file (checked 2026-10-21).
- Star counts differ by directory: truefoundry shows 43 and awesomeskill shows 2. GitHub's count is the one cited here.
- Copies: 306 exact copies of the SKILL.md exist in other repositories.
Common questions
Is Beautiful Prose free to use?
Yes, it is public and costs nothing to install. The repository has no licence file (GitHub, 2026-10-21), so reuse rights are unclear.
Does Beautiful Prose remove em dashes?
Yes, it bans them, but the rule shows only the double-hyphen form. The em-dash character itself never appears in the file.
Will Beautiful Prose keep my facts and numbers?
No, not by rule. Diff the claims after every rewrite.
What is the default register in Beautiful Prose?
The default is literary_modern. It applies when you set no REGISTER tag.
Does Beautiful Prose work in other languages?
Its rules target English prose. One adopter used it on app copy in Korean and English.
What writing does Beautiful Prose suit?
It suits essays, memos and narrative copy. It fits resumes, bulleted docs and LaTeX poorly.