What does this skill do, and when should you use it?
This is a pure-instruction Claude Skill: SKILL.md defines a binding style contract for English prose. It bans em dashes, "not X but Y" reversals, filler transitions, therapeutic tone, and AI meta-commentary, while prescribing declarative sentences, concrete nouns, strong verbs, and deliberate rhythm. Four registers plus density, heat, and length tags let you tune the voice. There are no scripts, tools, or dependencies; it works anywhere Agent Skills are supported. Caveats: license is unknown, and the referenced test-cases file's contents are not shown in the source.
Once activated, it constrains the model's English output: no em dashes, reversal pivots, banned filler phrases, validation language, or meta talk; mandatory declarative sentences, concrete nouns, strong verbs, and varied sentence length. Optional REGISTER/DENSITY/HEAT/LENGTH control lines tune register and intensity. Output is plain-text prose with no headings or bullets unless requested. It runs no scripts and calls no tools; it changes writing behavior only.
- Writers who want AI-drafted or rewritten English free of "At its core" and "That said" template cadence
- Content creators stripping therapeutic validation lines and empty summary paragraphs from LLM drafts
- Technical authors who want crisp, verb-forward English instead of content-marketing voice
- Users needing a fixed house voice, switching between the four registers via the REGISTER tag
- Editors using the banned-phrase list as a manual lint checklist when reviewing copy
How do you install this skill?
- License metadata is unknown; confirm terms before enterprise use.
- The referenced references/test-cases.md is absent from evidence, so the quality checklist cannot be statically verified.
- Skill targets English prose only; no Chinese support is declared, limiting fit for Chinese-language scenarios.
- Prompt-only constraints: actual de-AI-cadence effect depends on model execution; run small samples before adopting.
- Publisher is unverified and revision integrity is unguaranteed; pin the commit hash when using.
Installation steps are not documented in the repo. Following standard Agent Skills practice, clone the repository into your Claude Code skills directory (e.g. ~/.claude/skills/); the entry point is SKILL.md. License is unknown, so confirm terms before commercial use.
git clone --depth 1 https://github.com/SHADOWPR0/beautiful_prose.git ~/.claude/skills/beautiful_proseGenerated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Apply the Beautiful Prose skill.
Prepend your request with the line "Apply the Beautiful Prose skill."; the model should produce prose only, without acknowledging the skill. Optionally add control lines first, e.g. REGISTER: literary_modern / DENSITY: dense / HEAT: cool / LENGTH: long, then state the writing task. Without a REGISTER tag it defaults to literary_modern.
What are this skill's strengths and limitations?
- Highly specific ban list (reversal patterns, named filler phrases) that is directly actionable
- Tunable via four control tags: register, density, heat, length
- Pure Markdown instruction with zero scripts or dependencies; portable to any Agent Skills client
- Includes good/bad examples and a six-point manual lint checklist for QA
- English prose only; no effect on writing in other languages
- Compliance relies on the model's self-discipline; no automated lint script enforces the rules
- references/test-cases.md is referenced but its content is not shown, so test coverage is unclear
- No license declared and no README; maintenance status and author intent are unclear
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Beautiful Prose Style Contract this page | 50 · Use with care | ★ 57 | 9mo ago | — |
| No AI Slop | 59 · Recommended | ★ 12k | 1mo ago | MIT |
| Unslop | 49 · Use with care | ★ 10k | 11d ago | Apache-2.0 |
| e2e Docs Authoring Skill | 54 · Use with care | ★ 10k | 11d ago | Apache-2.0 |
| Khazix WeChat Long-Form Writer Skill | 49 · Use with care | ★ 21k | 24d ago | MIT |
How did FollowSkills review this skill?
Pure prompt-level style contract: no scripts, no external calls, no permission requests, no sensitive-data or destructive defaults; no red-line risk visible statically. Deducted for: attribution rests on an unverified publisher, license metadata is unknown, and there is no rollback or integrity mechanism confirming future revisions stay inert.
As an instruction-only skill it is internally consistent: banned lists, positive constraints, examples and lint checklist align, and activation syntax is explicit. Deducted for: no failure-feedback or abnormal-input behavior defined; `references/test-cases.md` is referenced but absent from the evidence, so the key verification path is not reproducible.
Trigger description is precise (English prose/rewrites free of AI tics), and REGISTER/DENSITY/HEAT/LENGTH tags give clear parameters; some non-fit scope is implicit. Deducted for: no Chinese-language support and no mention of environment fit for Chinese users; boundary conditions for academic or technical writing remain undefined.
Good layered structure (purpose, activation, prohibitions, constraints, registers, examples, lint, tests) with reasonable progressive disclosure. Deducted for: no license, no versioning or changelog, no maintainer or update path declared; the referenced test file is missing from evidence, leaving hidden assumptions.
The core value claim (eliminating AI cadence in English prose) cannot be verified statically; bad/good examples indicate intent but no representative outputs exist, and marginal value over a hand-written style prompt is unevidenced. Deducted for: entirely model-dependent outcomes with no comparative-benefit evidence.
Concrete banned-phrase lists and a manually checkable lint checklist are auditable primary material. Deducted for: no third-party execution evidence, no CI or committed tests covering key paths, the missing referenced test-cases file, and thin cross-source corroboration, leaving most claims as author assertions.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →