사용 방법
- 설치 명령을 복사하고 자신의 에이전트 환경에 설치하기 전에 소스 패키지를 확인하세요.
- 필요한 도구, 인증 정보, 지원 환경은 원본 문서에서 확인하세요.
- 입력과 원하는 산출물을 설명하고 출처와 결과를 대조하세요.
npx skills add https://github.com/d1zzy4/agent-skills --skill copywriting-expert시작 전 확인
- 설치에는 모델 구독, API 한도, 파일 접근 권한이 포함되지 않습니다.
- 별점 수와 설치 수는 활동 지표이며 독립적인 품질·보안 검증이 아닙니다.
원본 저장소의 제목, 예시, 코드, 표, 링크, 이미지를 유지해 표시합니다.
Copywriting Expert
Purpose
Write, audit, or improve user-facing product and UI copy across surfaces: buttons, labels, empty states, errors, tooltips, dialogs, toasts, onboarding, accessibility text, and CLI output. Check project-specific content guidance first and adapt to the product's language and audience rather than imposing a generic voice. Load only the component-specific reference that matches the problem.
Step 0: Establish the source of truth
Read references/project-source-of-truth.md. Project content standards, legal requirements, terminology, localization rules, and design-system guidance override portable defaults.
Read a sample of the project's existing user-facing copy before writing: UI strings from the codebase, the README, and any docs. These are the de facto house voice and terminology when no style guide exists.
If no source of truth exists, infer only from nearby product copy and explicit user requirements. Do not invent brand claims, policy promises, accessibility behavior, or legal guarantees.
Step 1: Identify the job of the copy
Before polishing wording, identify:
- What the user needs to understand.
- What action, if any, they can take.
- What can go wrong or be lost.
- Who the audience is and what language/register they use.
- Whether the copy is transactional, instructional, persuasive, or safety-critical.
Read references/voice-and-tone.md and the relevant component reference. For terminal surfaces (help text, flag descriptions, deprecation warnings, and non-interactive errors), read references/cli-output-copy.md.
Step 2: Write for comprehension first
Prefer specific verbs, plain language, active voice, useful nouns, and sentence case unless the product standard says otherwise. Keep the primary action obvious. Avoid jokes, euphemisms, and cleverness when they obscure consequences or increase cognitive load.
For destructive or irreversible actions, name the affected object and meaningful consequence. For errors, explain the problem and next step when a next step exists.
Step 3: Treat accessibility and localization as requirements
Read references/accessibility-and-localization.md when relevant. Do not rely on color, word length, capitalization, or idiom alone to communicate meaning. Avoid strings that become misleading when translated, pluralized, expanded, or rendered in a right-to-left locale.
Step 4: Verify terminology
Read references/language-and-vocabulary-verification.md when terminology or translation matters. Product names, technical terms, legal wording, and localized UI labels should come from authoritative sources, not guesswork.
Step 5: Run the final audit
Read references/examples-and-anti-patterns.md and references/formatting-and-punctuation.md. Check consistency across the whole flow, not just the changed string.
Anti-patterns
- Generic labels such as "Submit" when the actual action can be named.
- Errors that blame the user or expose raw implementation details.
- Confirmation dialogs for routine, reversible actions.
- Placeholder copy that ships.
- Unverified translations or product terminology.
- Generic AI filler ("delve", "leverage", "seamless") and essay signposting
("It's important to note that").
- Promising outcomes the product cannot guarantee.
- Writing a friendly tone that trivializes a high-stakes action.
- Em dashes in generated copy.
Bundled references
Load only the component-specific references needed for the task:
references/project-source-of-truth.md: checking for and deferring to a project's own content
style guide.
references/voice-and-tone.md: core voice principles and how tone shifts with stakes.references/ui-component-copy.md: buttons, labels, tooltips, form text.references/error-messages.md: how to write an error message that actually helps.references/empty-states.md: what an empty state needs to do beyond saying "nothing here".references/confirmation-dialogs.md: confirmation and destructive-action copy.references/toasts-and-onboarding.md: success feedback, undo actions, and first-run guidance.references/cli-output-copy.md: help text, flag descriptions, deprecation warnings, and
non-interactive terminal errors.
references/accessibility-and-localization.md: accessible names, status copy, and localization
constraints.
references/language-and-vocabulary-verification.md: verifying word choice and grammar against
authoritative per-language sources instead of guessing.
references/formatting-and-punctuation.md: the em dash ban and other punctuation rules.references/examples-and-anti-patterns.md: worked good/bad examples across component types.references/verification-and-failure.md: shared verification and failure-handling principles.references/proactive-trigger.md: when to review, rewrite, or audit copy without being asked,
and when to stay quiet.
