这是一组技能、系统提示和模板，面向AI原生构建者——产品经理、创始人和运营者，他们将LLM作为日常思考、交付和运行工作的工具。

这里的每个文件都设计成可以直接插入 Claude、ChatGPT、Cursor 或任何支持技能、系统提示或自定义指令的代理运行时。它们有主见，随时可用。

里面是什么

产品工作

• PRD-skill.md — 结合客户洞察和工程背景起草产品需求文档。根据综合输出读取（如有）。
• PRD-template.md — draft-prd使用的PRD结构。编辑标题，保持纪律。
• synthesize-customer-interview.md — 将原始访谈记录转化为具有引文级可追溯性、模式阈值和揭示矛盾的结构化发现。
• EVIDENCE_RULES.md — 综合过程中选择和引用客户报价的标准。
• JTBD_FRAMEWORK.md — 何时（以及何时不）应用待完成任务框架。
• OUTPUT_TEMPLATE.md — 访谈综合的固定输出结构。结构即学科。

参谋长特工

AI Builder Skills

A collection of skills, system prompts, and templates for AI-native builders — product managers, founders, and operators who use LLMs as a daily tool for thinking, shipping, and running their work.

Each file here is meant to be dropped into Claude, ChatGPT, Cursor, or any agent runtime that supports skills, system prompts, or custom instructions. They are opinionated and ready to use.

What’s inside

Product work

• PRD-skill.md — Draft a product requirements document by combining customer insights and engineering context. Reads from synthesis outputs when available.
• PRD-template.md — The PRD structure used by draft-prd. Edit the headings, keep the discipline.
• synthesize-customer-interview.md — Turn raw interview transcripts into structured findings with quote-level traceability, pattern thresholds, and surfaced contradictions.
• EVIDENCE_RULES.md — Criteria for selecting and citing customer quotes during synthesis.
• JTBD_FRAMEWORK.md — When (and when not) to apply Jobs-to-be-Done framing.
• OUTPUT_TEMPLATE.md — The fixed output structure for interview synthesis. The structure is the discipline.

Chief of Staff agent

A personal AI advisor for navigating a new senior role — strategy, execution, people, politics — grounded in a Context KB (MCP-backed).

• chief-of-staff-agent-design-doc.md — Full design doc for the agent: KB schema, MCP tools, behaviors, evaluation.
• chief-of-staff-system-prompt.md — The system prompt that runs the agent.
• morning chief of staff.md — Morning brief generator (calendar + email).
• end of day brief.md — End-of-day wrap-up across Calendar, Gmail, Drive, Jira.
• daily jira briefing.md — Daily sprint standup brief via the Atlassian connector.

How to use these

1. Skills (files with frontmatter — name: and description:) plug into Claude Code or any runtime that loads skill folders. Drop them in your skills directory.
2. System prompts can be pasted into a custom GPT, Claude Project, or any agent’s system slot.
3. Templates and frameworks are reference material — the skills above point to them as supporting context.

Conventions

• Skills are opinionated. They tell the model what not to do as often as what to do.
• Output structure is treated as the discipline. Don’t let the model reorder it.
• Evidence is traceable. Every claim should be linkable back to a quote, ticket, or source.

Contributing

PRs welcome. Keep skills:

• Single-purpose — one skill, one job.
• Self-contained — declare the inputs it reads and the outputs it writes.