测试审核记录保存4

在任何创造性工作之前必须使用 - 创建功能、构建组件、添加新功能或修改行为。通过协作对话探索用户意图、需求和设计。

Security Scan

This is an instruction-only brainstorming/design skill that is internally consistent: it asks the agent to inspect project context, produce design text, and (optionally) write a design file and commit it to git — nothing in the skill requests unrelated credentials or installs.

Purpose & Capability

The name and description describe a collaborative brainstorming/design helper; the SKILL.md content matches that purpose (questions to understand intent, propose options, produce 200–300 word design sections). There are no unrelated dependencies or requested credentials.

Instruction Scope

The instructions explicitly tell the agent to '查看当前项目状态（文件、文档、最近提交）' and to write validated designs to docs/plans/... and commit them to git. Asking to read the repository and commit files is coherent for this skill, but it implies the agent will access project files and perform write/git actions — verify you want the agent to have that filesystem/repo access.

Install Mechanism

No install spec and no code files: lowest-risk model. The skill is instruction-only and does not pull external binaries or archives.

Credentials

The skill declares no environment variables or credentials. This is proportional to a purely instructional skill; however, practical execution (committing to git) will rely on the agent runtime having git and the user's existing repo credentials configured — those are not requested by the skill itself.

Persistence & Privilege

always is false and the skill is user-invocable. It does not request permanent presence or attempt to modify other skills or global agent settings.

Assessment

This skill is an instruction-only brainstorming helper and appears coherent, but note two practical points before installing: - The SKILL.md asks the agent to inspect the current project (files, docs, recent commits). If you install this, ensure the agent runtime has only the repository and filesystem access you intend to expose — it will read project files to form context. - The skill also directs writing validated design docs to docs/plans/... and committing them to git. Committing uses whatever git credentials/config are present in the runtime. If you don't want automatic commits, restrict the agent's write or git access or confirm changes before committing. Also review any referenced sub-skills (core-writing-plans, embedded-* etc.) before allowing the agent to invoke them, and consider testing the skill in a sandboxed repository first.

2026年4月16日

Brainstorming - 将想法转化为设计

概述

通过自然的协作对话，帮助将模糊的想法转化为完整的设计和规格说明。

首先了解当前项目上下文，然后逐个提问来完善想法。一旦理解了要构建的内容，以小节形式（200-300字）呈现设计，每节后确认是否正确。

流程

理解想法

• 首先查看当前项目状态（文件、文档、最近提交）
• 逐个提问来完善想法
• 尽可能使用选择题，开放题也可以
• 每条消息只问一个问题 - 如果需要更多探索，分多个问题
• 聚焦于理解：目的、约束、成功标准

探索方案

• 提出 2-3 种不同方案及其权衡
• 以对话方式呈现选项，包括推荐和理由
• 优先展示推荐方案并解释原因

呈现设计

• 确信理解了要构建的内容后，呈现设计
• 分成 200-300 字的小节
• 每节后询问是否正确
• 涵盖：架构、组件、数据流、错误处理、测试
• 准备好返回澄清不清楚的地方

设计完成后

文档输出

• 将验证通过的设计写入 docs/plans/YYYY-MM-DD-<topic>-design.md
• 提交设计文档到 git

继续实施（如果需要）

• 询问：「准备好开始实施了吗？」
• REQUIRED SUB-SKILL: 使用 core-writing-plans 创建详细实施计划

领域扩展钩子

如果是嵌入式项目：

• RECOMMENDED: 使用 embedded-datasheet-analysis 分析硬件数据手册
• RECOMMENDED: 使用 embedded-feasibility-assessment 评估硬件约束
• RECOMMENDED: 使用 embedded-platform-selection 选择合适平台

如果是仿真项目：

• RECOMMENDED: 使用 simulation-requirements 分析仿真需求
• RECOMMENDED: 使用 simulation-architecture 设计仿真架构

核心原则

• 一次一个问题 - 不要用多个问题压垮用户
• 首选选择题 - 比开放题更容易回答
• YAGNI 原则 - 从所有设计中移除不必要的功能
• 探索替代方案 - 在确定之前始终提出 2-3 种方案
• 增量验证 - 分节呈现设计，验证每节
• 保持灵活 - 当某些内容不清楚时返回澄清