Skip to main content

create-prd

This skill should be used when the user asks to "create PRD", "write product requirements document", or mentions "PRD", "product requirements document".

ソース情報

リポジトリ
FradSer/dotclaude
ソースの最終更新活動
2026年8月8日 08:26
検出された SKILL.md の言語
英語
スター
593
フォーク
50

インストール方法

デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。

ソースファイルを確認

インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。

ファイルエクスプローラー
8 ファイル

SKILL.md を表示中

SKILL.md
ソースの指示 · 読み取り専用プレビュー
name
create-prd
description
This skill should be used when the user asks to "create PRD", "write product requirements document", or mentions "PRD", "product requirements document".
user-invocable
true
argument-hint
--md | --lark [--folder-token TOKEN_OR_URL | --wiki-node TOKEN_OR_URL | --wiki-space ID_OR_URL]
allowed-tools
["Read","Write","AskUserQuestion","Glob","Bash(lark-cli:*)","Skill"]
# PRD Creation Follow the phases below to transform product ideas into a complete PRD document. All PRD content is written in Chinese (the target audience is Chinese-speaking teams). ## Output Mode Determine output mode from `$ARGUMENTS`: | Argument | Mode | Output | |----------|------|--------| | `--md` (default) | Local Markdown | Save as `.md` file to project directory | | `--lark` | Feishu Cloud Doc | Create via `lark-cli` using Lark rich-text features | `--lark` mode accepts optional location arguments (mutually exclusive), supporting token or Feishu URL: - `--folder-token` — Target folder (token or URL) - `--wiki-node` — Target wiki node (token or URL) - `--wiki-space` — Target wiki space root (ID or URL, or `my_library`) Pass URL directly to `lark-cli` — no manual token extraction needed. Defaults to user's personal space root when no location specified. ## Phase 0: Import Context Check for existing design or requirements documents: 1. Search for `docs/`, `prd/` directories or user-specified files 2. If found, extract key information (problem statement, target users, core features) as pre-filled content 3. If nothing found, skip this phase ## Phase 1: Determine PRD Type Ask the user which PRD type they need: - **Full** (recommended): All standard sections, suitable for complex features - **Brief**: Core sections only, suitable for small features or quick iteration - **One-pager**: Single-page summary, suitable for concept validation and executive reporting ## Phase 2: Gather Information Follow the interview questions in `references/prd-interview-questions.md`, one question at a time. Wait for the answer before continuing. - Basic info (7 items): required for all types - Full version extras (5 items): full type only - AI Agent boundaries: needed when the PRD will be consumed by AI coding agents If Phase 0 found pre-filled content, show it to the user for confirmation and skip covered questions. ## Phase 3: Generate PRD Document 1. **Select template**: - Full: `references/prd-template-full.md` - Brief: `references/prd-template-brief.md` - One-pager: `references/prd-template-onepager.md` 2. **Fill content**: Use gathered information for each section, following `references/prd-best-practices.md` writing principles 3. **AI Agent consumability**: - Write each requirement as a discrete, verifiable item (lists over long paragraphs) - Express non-goals as positive constraints - Split P0 features into 5-15 minute agent work stages with testable checkpoints - Full version includes a three-layer boundary framework: autonomous / needs confirmation / prohibited 4. **Quality requirements**: - Problem statement backed by specific data or research - Goals follow SMART principles - Success metrics are quantifiable - Avoid vague terms ("approximately", "maybe", "try to") - Use active voice and concrete verbs 5. **Format**: - `--md` mode: Standard Markdown with clear heading hierarchy - `--lark` mode: Lark-flavored Markdown (see Feishu enhancements section) ## Phase 4: Validate and Save Run validation per `references/prd-validation-checklist.md` — completeness, SMART goals, content quality, BDD acceptance criteria. ### `--md` Mode Save file after validation: - Filename: `PRD-[ProductName]-[YYYYMMDD].md` - Prefer `docs/` or `prd/` directory, otherwise current working directory - Report path and file summary ### `--lark` Mode Create Feishu document after validation: 1. **CRITICAL** — Confirm standalone `lark` plugin (`lark@frad-dotclaude`) is installed; follow its `lark-shared` skill for authentication 2. Follow the lark plugin's `lark-doc` skill for document creation guidance 3. Refer to `lark-doc-create.md` for full `docs +create` parameters and Lark-flavored Markdown syntax 4. Convert PRD content to Lark-flavored Markdown (see Feishu enhancements section) 5. Create document: ```bash lark-cli docs +create --title "PRD-[ProductName]-[YYYYMMDD]" \ [--folder-token TOKEN_OR_URL | --wiki-node TOKEN_OR_URL | --wiki-space ID_OR_URL] \ --markdown "<lark-flavored-markdown>" ``` 6. For longer PRDs, split: `docs +create` for first half, then `docs +update --mode append` for remaining sections 7. If `board_tokens` are present: - Follow the lark plugin's `lark-whiteboard` skill - Fill each whiteboard with actual content (flowcharts, architecture diagrams) - All whiteboards must have real content before task is complete 8. Report the document URL ## Phase 5: Next Steps After saving, suggest follow-up options: - Convert PRD into implementation tickets - Refine design decisions in open questions - Share with team for review and feedback ## Feishu Document Enhancements **CRITICAL** — In `--lark` mode, use Lark-flavored Markdown syntax to leverage Feishu's rich-text capabilities. ### Required Feishu Features by PRD Section | PRD Section | Feishu Feature | Description | |---|---|---| | Project metadata (version/date/owner) | `<lark-table>` | Enhanced table with header row | | Key risks/assumptions/dependencies | `<callout>` | Color-coded: risk=red, assumption=blue, dependency=yellow | | Core goals / success metrics | `<callout emoji="..." background-color="light-green">` | Highlight key OKRs | | Priority comparison (P0/P1/P2) | `<grid cols="3">` | Three-column side-by-side layout | | User journey / business process | `<whiteboard type="blank">` | Flowchart, filled via lark-whiteboard | | System architecture | `<whiteboard type="blank">` | Architecture diagram, filled via lark-whiteboard | | Milestones / timeline | `<whiteboard type="blank">` | Timeline chart | | Non-goals / scope exclusions | `<callout emoji="..." background-color="light-red">` | Red highlight for prohibited scope | | Glossary / abbreviations | Two-column Markdown table | Compact reference | | Decision records | `<callout>` + blockquote | Key decisions with rationale | ### Whiteboard Rules Insert whiteboards for: user journeys, system architecture, data flows, milestones, team organization. Skip for pure text, data-heavy content (use tables), or when user requests text-only. ### Format Principles - Max 4 heading levels - Use `---` dividers between sections - Do NOT write a top-level heading duplicating the title (Feishu auto-generates it) - Use callouts sparingly; bold only core terms - Use `<text color="red">` for key metrics or status - Feishu auto-generates table of contents — do not add manually ## Quality Principles - **Data-driven**: Support problem statements with specific data and user research - **SMART goals**: Specific, Measurable, Achievable, Relevant, Time-bound - **Concise and clear**: Avoid verbosity; descriptions must be clear enough for dev teams to implement directly - **Collaboration-oriented**: PRD is a collaboration tool; tone promotes discussion, not command - **Dual-audience design**: Serve both human teams and AI coding agents ## Supporting Files - `references/prd-interview-questions.md` — Information gathering questionnaire - `references/prd-validation-checklist.md` — Validation checklist - `references/prd-template-full.md` — Full version template - `references/prd-template-brief.md` — Brief version template - `references/prd-template-onepager.md` — One-pager template - `references/prd-best-practices.md` — Best practices guide - `references/prd-examples.md` — High-quality PRD examples - Standalone `lark` plugin (`lark@frad-dotclaude`) — Lark CLI skills (`--lark` mode)
GitHubで見る