跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
安全

prd-generator

Generate a full security-focused PRD (Product Requirements Document) from a plain-language feature description. Use this skill whenever someone wants to create a PRD, product spec, feature spec, requirements document, or says things like "write a PRD for", "create a spec for", "plan this feature", "document this feature", or "I need a requirements doc". The skill produces: (1) a structured PRD in a security-focused template format, (2) a Confluence page pushed to your configured space via Atlassian MCP (called directly), (3) a cursor-compatible .md plan file for .cursor/plans/, (4) a downloadable formatted markdown, (5) optionally: Linear/Asana tasks from the batch plan, Slack notification to stakeholders, Notion page mirror. Always use this skill for PRD and feature planning tasks.

Security-Phoenix-demo · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add Security-Phoenix-demo/security-skills-claude-code --skill prd-generator

技能说明


name: prd-generator description: > Generate a full security-focused PRD (Product Requirements Document) from a plain-language feature description. Use this skill whenever someone wants to create a PRD, product spec, feature spec, requirements document, or says things like "write a PRD for", "create a spec for", "plan this feature", "document this feature", or "I need a requirements doc". The skill produces: (1) a structured PRD in a security-focused template format, (2) a Confluence page pushed to your configured space via Atlassian MCP (called directly), (3) a cursor-compatible .md plan file for .cursor/plans/, (4) a downloadable formatted markdown, (5) optionally: Linear/Asana tasks from the batch plan, Slack notification to stakeholders, Notion page mirror. Always use this skill for PRD and feature planning tasks.

PRD Generator — Security-Focused

What this skill can talk to

None of these are required — the PRD is produced either way, and a missing connector degrades to a local file rather than an error.

MCP connectorUsed forPriority
AtlassianConfluence page creationPrimary — always attempt
SlackStakeholder notificationOptional
LinearBatch plan → ticketsOptional
AsanaBatch plan → tasksOptional
NotionMirror docOptional
GmailSend the PRD to stakeholdersOptional

Tools it expects: present_files, create_file, bash_tool.

This list used to live in a compatibility: frontmatter key. The Agent Skills spec allows only name, description and allowed-tools, and any extra key makes the claude.ai upload fail with a hard error instead of being ignored — so it belongs in the body.

Customization (Required on First Use)

Before generating your first PRD, Claude will ask you to provide the following settings. These are used throughout the pipeline and in all outputs.

SettingDescriptionExample
Owner nameName(s) shown as PRD owner in all outputs@jane.doe / @john.smith
Default stakeholdersInterested parties added to every PRD@product-team, @security-lead
Confluence space keyYour Atlassian Confluence space for PRD pagesENG, PRODUCT, SEC
Confluence template titleParent page title to nest PRDs underPRD Templates, Product Specs
Slack channelDefault channel for PRD notifications (optional)#product-specs, #security-reviews
Email recipientsDefault stakeholder emails for Gmail drafts (optional)team@example.com

How to configure:

  1. Interactive (recommended): The first time you use this skill, Claude will ask you for these values before generating the PRD. Your answers are used for that session.

  2. Environment variables: For persistent configuration, set these in your shell profile or .env:

    export PRD_OWNER="@your-name"
    export PRD_STAKEHOLDERS="@stakeholder1, @stakeholder2"
    export PRD_CONFLUENCE_SPACE="YOUR_SPACE_KEY"
    export PRD_CONFLUENCE_TEMPLATE="Your Template Page Title"
    export PRD_SLACK_CHANNEL="#your-channel"
    
  3. Per-request override: Include any of these in your prompt:

    Write a PRD for feature X. Owner: @jane. Space: ENG. Notify #product-specs.
    

If no configuration is provided, Claude will prompt you before proceeding.


What This Skill Does

Takes a feature description → runs the 10-role spec pipeline → produces all outputs. Connectors are called directly by Claude — no separate user action required.

OutputConnector / Tool
Full PRD Markdowncreate_file + present_files
.cursor/plans/ filecreate_file + present_files
Confluence page in configured spaceAtlassian:createConfluencePage (called directly)
Linear issues from batch planLinear MCP (if user confirms)
Asana tasks from batch planAsana MCP (if user confirms)
Slack notificationSlack:slack_send_message (if channel provided)
Notion mirrorNotion:notion-create-pages (if user confirms)
Email to stakeholdersGmail:gmail_create_draft (if user confirms)

Execution Steps

Step 1 — Gather Configuration + Extract Intent

1a. Check for customization settings. Look for:

  • Environment variables (PRD_OWNER, PRD_STAKEHOLDERS, PRD_CONFLUENCE_SPACE, etc.)
  • Values provided in the user's prompt (e.g., "Owner: @jane", "Space: ENG")
  • If neither is available, ask the user before proceeding:

    "Before I generate the PRD, I need a few details:

    1. Who should be listed as owner? (e.g., @your-name)
    2. Who are the key stakeholders? (e.g., @product-lead, @security-lead)
    3. What Confluence space should I publish to? (e.g., ENG, PRODUCT)
    4. Should I notify a Slack channel or email anyone? (optional)"

1b. Extract intent from the user's description:

  • Feature name → becomes PROJECT-NAME in the Confluence title
  • Problem statement — why this exists
  • Rough scope — what's in / out
  • Which connectors to activate — infer from context; see heuristics at bottom

Ask at most one clarifying question if critical info is missing. Then proceed.


Step 2 — Run the Spec Pipeline (Roles 01–10, inline)

Execute each role mentally. Chain outputs:

RoleOutputToken Budget
01 Context CuratorCLEAN_CONTEXT: facts, decisions, open Qs≤900
02 Scope CutterSCOPE_DEFINITION: goals, non-goals, in/out≤800
03 Constraint DistillerACTIVE_SET: 8–15 testable ACs≤700
04 Requirements EngineerNORMATIVE_REQUIREMENTS: RFC-2119, IDs, P0/P1/P2≤1500
05 Ambiguity HunterCLARIFICATIONS: flag critical ambiguities; loop back to 04 if needed (max 3x)≤1000
06 Security EngineerSECURITY_REQUIREMENTS + THREAT_MODEL≤1200
07 Contract ArchitectCONTRACTS: APIs, events, er
阅读完整来源 ↗