公司内部系统PRD编写

Use this skill when the user wants to design a brand-new (0→1) PRD or prototype for a company-internal process management system, where roles and workflows a...

Security Scan

The skill's instructions, file layout, and resource needs are consistent with its stated purpose (0→1 PRD + prototype for internal process systems); it requests no credentials, no installs, and contains no unexpected external endpoints.

Purpose & Capability

Name/description claim to produce PRD, workflows, RBAC, tracking plan and an HTML prototype — the SKILL.md and reference files describe exactly those outputs. There are no environment variables, binaries, or install steps that would be unrelated to producing documents and an HTML prototype.

Instruction Scope

Instructions stay focused on product-design tasks (input collection, validations, user checkpoints, ER/flows/RBAC/tracking and a single-file HTML prototype). One notable instruction is domain-shift detection: Step 2.1 requires scanning the session/conversation history to detect field changes. This is relevant to the task but means the skill will read prior conversation context; users should be aware that previous messages may be used as input. The skill does not instruct exporting data to external endpoints, but the prototype spec uses Tailwind CDN (a public CDN) for styling — browser fetches to the CDN are expected when viewing the generated HTML.

Install Mechanism

Instruction-only skill with no install spec and no code files; this minimizes on-disk installation risk. The prototype is produced as single HTML using Tailwind CDN and native JS (no packaged installers or downloads).

Credentials

No environment variables, credentials, or config paths are requested. All required inputs are collected interactively from the user (roles, pain points, expectations, decisions). This is proportionate to the stated functionality.

Persistence & Privilege

always:false and standard autonomous invocation apply. The skill does not request persistent presence or modify other skills or system-wide settings. It writes structured artifacts under a local project/ layout (as part of its workflow), which is expected for a document/prototype generator.

Assessment

This skill appears coherent and focused on producing internal PRDs and a single-file prototype, but consider these practical points before installing: - Confirm you're comfortable with the agent reading prior conversation history (Step 2.1) because it uses that context to detect domain shifts. - Avoid supplying production secrets or PII when authoring input; generated prototype HTML may include sample data and will reference Tailwind CDN (viewing the HTML causes a browser to fetch that CDN). - The skill intentionally refuses to invent API/interface details and punts technical design to a separate tech-design skill — if you need API specs, plan to run that downstream skill. - Review any generated prototype HTML before deploying or sharing; ensure it does not embed real credentials, analytics keys, or third‑party scripts beyond the stated Tailwind CDN. If you want higher assurance, request the author to confirm: (a) whether the agent will persist the project/ files to a shared location, and (b) whether any telemetry or analytics are enabled when the prototype is rendered.

Product Design 0→1 (Internal Process System) — v2.1

7-step pipeline. Each step has a human checkpoint. Latest confirmed artifact flows to next step.

Domain assumption

Target is ALWAYS a company-internal process management system. Roles and workflows are core. NOT for ToC products.

File layout (under project/)

project/
├── 00-input/{overview, roles, pain-points, expectations}.md
├── 01-user-stories.md
├── 02-flows/{swimlane.md, state-machine.md}
├── 03-er-model.md            # 🆕 v2.1
├── 03.5-field-decisions.md
├── 04-modules.md
├── 05-rbac-matrix.md         # 🆕 v2.1
├── 06-tracking-plan.md       # 🆕 v2.1
└── 07-output/{prototype.html, prd.md}

The 7 steps (each delegated to a reference file)

Step	Topic	Reference
1	入口确认	inline (one sentence)
2	输入收集（4 子步 + 领域偏移检测）	references/steps/step2-input-collection.md
3	7 条机械校验	references/steps/step3-validation.md
4	用户故事（三段式 + AC）	references/steps/step4-user-stories.md
5	泳道图 + 状态机	references/steps/step5-flows.md
5.5	字段决策清单	references/steps/step5.5-field-decisions.md
6	功能模块 + ER 模型 + 字段	references/steps/step6-modules-er.md
6.5	RBAC 权限矩阵 + 埋点规划	references/steps/step6.5-rbac-tracking.md
7	高保真原型 + 12 节 PRD	references/steps/step7-output.md

Cross-cutting rules (always apply)

1. 强制卡点：每步产物用 present_files 呈现后必须等用户确认（"确认"/"修改 X"），不允许越权进入下一步
2. AI 推断必须显式化：所有非用户明示的细节（字段、状态机、权限、埋点）必须通过 Step 5.5 + Step 6.5 让用户拍板
3. 回退即重生成：任何回退后，下游产物全部重做，禁止智能合并
4. [TODO:] 即时标注：当某节信息确实不足时，PRD 中用 [TODO: 具体需要补什么] 占位，绝不编造（吸收自 create-prd）
5. 领域偏移检测：Step 2.1 前必须扫描会话历史，发现领域切换主动暂停询问 a/b/c
6. 不发明：角色/痛点/字段/权限/埋点必须能从用户输入或决策清单溯源

Self-check

完成所有步骤后加载 references/appendices/selfcheck.md 跑自检清单，输出"通过/待补"清单。

End

"0→1 设计完成（含 PRD / 原型 / RBAC / 埋点）。下一步可调用 tech-design skill 进行技术方案设计。"