d1zzy4/agent-skills

copywriting-expert

Write, audit, or improve user-facing product and UI copy including buttons, labels, empty states, errors, tooltips, dialogs, toasts, onboarding, accessibility text, and CLI output such as help text, flags, deprecation warnings, and terminal errors.

ソースを見る

使い方

  1. インストールコマンドをコピーし、自分の環境へ導入する前にソースを確認します。
  2. 必要なツール、認証情報、対応環境を原文で確認します。
  3. 入力と成果物を指定し、出力を資料と照合します。
npx skills add https://github.com/d1zzy4/agent-skills --skill copywriting-expert

開始前の確認

  • 導入にはモデル契約、API枠、ファイルへのアクセス許可は含まれません。
  • スター数や導入数は活動の指標であり、独立した品質・安全性試験ではありません。
リポジトリの原文

見出し、例、コード、表、リンク、参照画像を含む原文を表示しています。

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.

タスクに関連する項目

同じリポジトリから

関連する Skills

すべての Skills
d1zzy4
コミュニティ

commit-expert

Guide safe, repository-aware Git commits and pushes. Inspect repository policy, working-tree state, diffs, hooks, tests, branch/upstream configuration, and commit conventions before any mutation. Never commit, push, stage, restore, or discard changes without explicit authorization for that side effect. Use Conventional Commits only when repository policy or the user requires it.

導入数
1
GitHub Stars
1
更新日
9月28日
d1zzy4
コミュニティ

create-readme-expert

Create comprehensive, well-structured README.md files for software projects. Trigger when a project needs a README, when onboarding documentation is missing, when a repository lacks clear usage instructions, or when README content is being reviewed or rewritten. Follow the project's own style and audience; do not impose generic README templates where the project already has strong conventions.

導入数
1
GitHub Stars
1
更新日
9月28日
d1zzy4
コミュニティ

mermaid-diagrams-expert

Create maintainable Mermaid diagrams for software documentation, including flowcharts, sequence, class, ER, C4, architecture, state, git, gantt, chart, kanban, packet, venn, and ishikawa diagrams. Trigger when structure, relationships, sequencing, or architecture would be clearer visually, especially for persistent README, wiki, PR, or design-document diagrams. Verify the target renderer and Mermaid version before using version-sensitive or beta syntax.

導入数
1
GitHub Stars
1
更新日
9月28日
d1zzy4
コミュニティ

redis-expert

All-in-one Redis expertise covering data modeling and key naming, client connections (pooling, pipelining, client-side caching, timeouts), clustering and replication, Redis Search (FT.CREATE, vector and hybrid search, RAG pipelines), observability and incident triage, production security hardening, semantic caching for LLM responses via LangCache, and agent memory via Redis Iris (session events, long-term memory, background promotion). Use whenever designing, writing, reviewing, or debugging any

導入数
1
GitHub Stars
1
更新日
9月28日