Token导航 LogoToken导航TokenDH.com
研究检索只读github未标认证来源可访问许可证需确认审计通过

write-prd写 prd

Agent Skill

write-prd 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

321

周安装

13

GitHub Stars

公开资料未说明

下载量

101
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

unknown

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:write-prd(写 prd)
来源仓库:https://github.com/queso/ai-team-skills
仓库路径:skills/write-prd
安装命令:
npx skills add https://github.com/queso/ai-team-skills --skill write-prd
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/queso/ai-team-skills --skill write-prd

简介

write-prd 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。

  • 适用于研究检索类任务,可结合来源仓库、安装命令和原始 README 继续核验具体用法。
  • 通过 npx skills add 命令从指定 GitHub 仓库安装,需确认权限范围和维护状态。
  • 安装前建议确认是否会触发联网、命令执行或文件读写,避免误操作生产数据。
  • write-prd 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

SKILL.md

Write PRD

Create a new Product Requirements Document in the project's PRD directory.

Step 1: Find the PRD Directory and Next Number

Look for an existing PRD directory in the project. Common locations: docs/prds/, prds/, docs/prd/, prd/. If none exists, ask the user where PRDs should live.

List existing PRD files and determine the next sequence number. Files follow the pattern NNNN-slug.md (e.g., 0019-user-onboarding.md). The next file would be 0020-<slug>.md. If no PRDs exist yet, start at 0001.

Step 2: Mini-Discovery Workshop

Before writing, run a structured discovery pass. There are four areas to cover:

  1. Problem & users — Who struggles today, and what breaks for them?
  2. Business value — What metric, revenue, or cost does this affect?
  3. Constraints — Tech stack, platform, timeline, must-nots?
  4. Risks — What's most likely to make this fail or get stuck?

How to run:

  1. Review what the user already provided. Read their input, any linked documents, and project context (CLAUDE.md, package.json, etc.).
  2. Identify gaps. Skip any question whose answer is already clear:

- Skip *Problem & users* if the user named a specific user/persona AND described a concrete pain point. - Skip *Business value* if the user connected the feature to a measurable business outcome. - Skip *Constraints* if the user stated constraints, OR this is a PRD for the current project (infer stack from CLAUDE.md/package.json). - Skip *Risks* if the user identified specific risks.

  1. Ask remaining gaps in ONE message. Batch all unanswered questions into a single ask — do not ask them one at a time. If all four areas are covered: *"You've given me strong context. Let me draft the PRD."*
  2. Max one round of follow-up. If answers are thin, fold those areas into "Open Questions" in the PRD rather than pressing further.
  3. Assess scale. Based on the scope of the feature, determine which tier to use: Tier When Required Sections Quick Bug fix, small enhancement, 1-2 day effort Problem Statement, Scope, Requirements Standard Feature, multi-day effort Sections 1-6 + 10 Deep New product area, multi-week, cross-team All 11 sections Tell the user which tier you'll use: *"This looks like a Standard-tier PRD — I'll include sections 1 through 6 plus Risks & Open Questions. Let me know if you'd prefer a different depth."* The user can override.

Step 3: Generate the Filename

Combine the next sequence number with 2-4 kebab-case keywords from the feature description:

  • 0020-fix-social-sharing.md
  • 0021-add-team-billing.md
  • 0022-migrate-auth-to-oauth2.md

Confirm the filename with the user before writing.

Step 4: Write the PRD

Use the template structure from references/prd-template.md. Include only the sections appropriate for the assessed tier. The PRD should be:

  • Problem-first — Lead with why, not what
  • Measurable — Every goal has a success metric
  • Scoped — Explicit about what's in and what's out
  • User-centered — Written from the user's perspective, not the developer's
  • Implementation-free — Describe what, not how (no architecture, no code)
  • Strategy-level Solution Approach — If included, keep Solution Approach at a level a product manager would understand. No code, no schemas, no class names.

Consult references/prd-best-practices.md for good and bad examples. Consult references/prd-template.md for the document structure.

Step 5: Review and Refine

After writing the first draft, review it against this checklist:

  • Does every requirement have a clear "why"?
  • Are success metrics specific and measurable?
  • Is scope clearly bounded (in/out)?
  • Are edge cases and error states folded under Requirements?
  • Are constraints and dependencies captured in Technical Considerations?
  • Are risks and open questions called out?
  • Does the Solution Approach (if present) stay at strategy level, free of implementation details?
  • Is the tier appropriate — not over-templated (a bug fix with 11 sections) or under-specified (a new product area with 3)?

Present the draft to the user for feedback before finalizing.

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

34.69%
按下载量换算35

Claude

29.74%
按下载量换算30

Cursor

19.5%
按下载量换算20

Gemini CLI

9.97%
按下载量换算10

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

只读

该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills