Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/Morrison-Lab/ai-config --skill use-preferred-style명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
SKILL.md 표시 중
| name | use-preferred-style |
| description | Apply preferred writing style. |
| user-invocable | true |
| allowed-tools | ["Bash","Read","Edit","Write"] |
Apply the user's prose style to anything user-facing. The style is plain and direct. Say the point in a short sentence. Stack fewer clauses. Drop filler.
The authority is the user's own guide, Principles of Scientific Writing (PSW): https://morrison-lab.github.io/psw/. PSW covers word choice, conciseness, grammar, and avoiding AI tells --- including the nested-dependent-clause rule below, in its own "Conciseness" chapter. This skill operationalizes PSW and adds the user's own rule on joining independent clauses with coordinating conjunctions rather than subordinate constructions, which PSW does not state explicitly. When in doubt, defer to PSW.
/style.Rule 1 restates PSW's own "Limit complex sentence structures" section (Conciseness chapter). Rules 2--4 are the user's own emphasis on filler and clause structure, drawing on PSW's conciseness and word-choice material without matching one specific section. Rules 5--8 come straight from PSW's word-choice, conciseness, and grammar chapters, cited individually below.
Limit dependent clauses. A dependent (subordinate) clause cannot stand alone. It starts with a word like because, although, which, while, since, when, or if. One per sentence is plenty. Two or more stacked together is a rewrite signal. Break the sentence apart instead.
Cut low-content filler and jargon. Delete words that add no information. Replace jargon with the plain word. See the swap table below.
Prefer simple declarative sentences. State the fact. Put the subject first, then the verb, then the rest. Short beats clever.
Join independent clauses with coordinating conjunctions. Two complete thoughts read better side by side than nested. Use and, but, so, or, yet, nor, for. Prefer "X is fast, but Y is correct" over "While X is fast, Y is correct."
Prefer plain (Anglish) words over Latin-derived ones (PSW, "Word choice"). Old-English-derived words decompose into parts a reader already knows. Latin roots must be memorized. So write before, not prior to; needed, not necessary; use, not utilize. This is a heuristic, not a purity rule — pick whatever word the reader grasps fastest.
Prefer active voice (PSW, "Conciseness"). Name the actor, then the action. Prefer "The researchers ran the experiment" over "The experiment was run by the researchers." Passive is fine when the actor is unknown or beside the point.
Keep relative pronouns (PSW, "Grammar"). English lets you drop that, which, or who in some sentences, but dropping it forces the reader to backtrack once they realize a describing clause has started. Prefer "the groundwork that I now apply" over "the groundwork I now apply." Also use that for a restrictive clause (identifies which thing you mean, no comma) and which for a non-restrictive one (adds a detail, set off by commas): "the dataset that we collected in 2023" versus "our dataset, which contains 4,000 records,".
Name a demonstrative's referent (PSW, "Grammar"). This, that, these, and those used standalone as pronouns send the reader searching backward for what they mean, often across a whole preceding clause rather than one noun. Follow the demonstrative with the noun it refers to.
| Cut / avoid | Use instead |
|---|---|
| in order to | to |
| due to the fact that | because |
| at this point in time | now |
| prior to | before |
| in the event that | if |
| necessary | needed |
| has the ability to, is able to | can |
| utilize, leverage | use |
| facilitate | help, let |
| make a decision | decide |
| give consideration to | consider |
| a large number of, a number of | many, some, N |
| it is worth noting that | (delete) |
| it should be noted that | (delete) |
| needless to say | (delete) |
| as a matter of fact | (delete) |
| basically, essentially, actually | (delete) |
| very, really, quite, simply | (delete) |
| in terms of | (rephrase or delete) |
| with regard to | about |
Treat the table as a starting set, not a closed list. Any word that carries no information is filler. Cut it. Most of these swaps come from PSW's "Conciseness" and "Word choice" chapters.
Before: It is worth noting that, due to the fact that the cache was not being invalidated when the user updated their profile, which caused stale data to be served, we decided to utilize a new eviction strategy in order to resolve the issue.
After: The cache was not invalidated on profile update. So it served stale data. We switched to a new eviction strategy to fix it.
The "after" is three sentences. Each makes one point. No filler, no utilize,
no stacked which/due to the fact that clauses.
This is a default, not an absolute rule. Keep a dependent clause when splitting it would read worse or lose a real logical link. Keep a technical term when it is the precise word and the plain swap would be wrong. Readability wins, not minimalism for its own sake. (Mirrors the "avoid nesting, but not blindly" stance in the coding-style rule.)
The user's guide, Principles of Scientific Writing, is the source of record. Pull the latest rules from it, not from this summary:
find-ai-tells skill below.PSW is a work in progress. When it and this skill disagree, PSW wins — and flag the drift so this skill gets updated.
find-ai-tells --- the detector counterpart.
It scans finished text for AI-authorship tells.
This skill prescribes how to write up front.
Run find-ai-tells after; run use-preferred-style during.simplify / tidy — the same "cut what adds no value" instinct, applied
to code instead of prose.memorize / remember — for a one-off wording preference, write a memory
instead of editing this skill.❌ The model overfit the training data. This poses a problem for deployment.
✅ The model overfit the training data. This overfitting poses a problem for deployment.