design-synthesis

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

[H1][DESIGN-SYNTHESIS]

Dictum: Design decisions require grounded context before implementation.

Synthesize research findings into design decisions via light codebase investigation.

Workflow:

§INGEST — Load research artifact, parse original request
§SCAN — Light codebase investigation via
```
parallel-dispatch
```
(3-4 agents)
§EXPLORE — Generate 2-3 approaches with trade-offs
§SELECT — Commit to best approach with rationale
§OUTPUT — Structured design document

Dependencies:

```
parallel-dispatch
```
— Agent orchestration for codebase scan
Research artifact — External findings from
```
deep-research
```

Input:

```
Research
```
: Path to research artifact (
```
research_{slug}.md
```
)
```
Request
```
: Original user request/intent

准则： 设计决策在落地前必须基于真实场景。

通过轻量代码库调研将研究成果转化为设计决策。

工作流程：

§数据摄入 — 加载研究成果文件，解析原始需求
§扫描 — 通过
```
parallel-dispatch
```
进行轻量代码库调研（3-4个Agent）
§探索 — 生成2-3种带有优劣分析的实现方案
§选择 — 确定最优方案并给出理由
§输出 — 结构化设计文档

依赖项：

```
parallel-dispatch
```
— 用于代码库扫描的Agent编排工具
Research artifact — 来自
```
deep-research
```
的外部研究成果

输入：

```
Research
```
: 研究成果文件路径（
```
research_{slug}.md
```
）
```
Request
```
: 原始用户需求/意图

[1][INGEST]

Dictum: Grounded context prevents speculative design.

Load and parse inputs:

[INDEX]	Source	Extract
[1]	Research file	Findings, confidence levels, key sources
[2]	Request	Intent, scope boundaries, success criteria

Parse research structure:

```
## [1][FINDINGS]
```
→ Domain knowledge by category
```
## [2][CONFIDENCE]
```
→ High/Medium/Low ratings
```
## [3][SOURCES]
```
→ Attribution for decisions

[IMPORTANT]:

[ALWAYS] Extract high-confidence findings as primary input.
[ALWAYS] Note low-confidence areas as design risks.
[NEVER] Proceed without understanding request intent.

准则： 基于真实场景的上下文可避免投机性设计。

加载并解析输入内容：

[序号]	来源	提取内容
[1]	研究文件	研究结论、置信度等级、关键来源
[2]	用户需求	意图、范围边界、成功标准

研究文件结构解析：

```
## [1][FINDINGS]
```
→ 按分类整理的领域知识
```
## [2][CONFIDENCE]
```
→ 高/中/低置信度评级
```
## [3][SOURCES]
```
→ 决策的参考来源

[重要提示]:

[必须] 将高置信度的研究结论作为主要输入。
[必须] 将低置信度的领域标记为设计风险。
[禁止] 在未理解需求意图的情况下推进工作。

[2][SCAN]

Dictum: Pattern awareness prevents reinvention.

Dispatch 3-4 agents via

parallel-dispatch

for codebase context.

Agent Assignment:

[INDEX]	[AGENT]	[SCOPE]	[RETURNS]
[1]	Patterns	Similar implementations in codebase	Conventions, reusable patterns, prior art
[2]	Constraints	Project rules, architecture limits	Hard boundaries, style requirements
[3]	Interfaces	Entry/exit points for feature area	Touch points, consumers, integration surface

Agent Prompt Template:

Scope: [Specific investigation area]
Objective: Surface [patterns|constraints|interfaces] relevant to: [request summary]
Output: Bullet list of findings with file paths
Context: Research indicates: [key findings summary]
Exclusions: Do NOT analyze implementation details or specific file contents

[CRITICAL]:

[ALWAYS] Dispatch ALL agents in ONE message block.
[ALWAYS] Scope to patterns/constraints/interfaces—NOT implementation.
[NEVER] Deep-dive into file contents—that's plan's job.

准则： 了解现有模式可避免重复造轮子。

通过

parallel-dispatch

调度3-4个Agent以获取代码库上下文。

Agent分配：

[序号]	[AGENT]	[范围]	[返回内容]
[1]	模式分析Agent	代码库中的类似实现	编码规范、可复用模式、已有解决方案
[2]	约束分析Agent	项目规则、架构限制	硬性边界、风格要求
[3]	接口分析Agent	功能模块的入口/出口点	交互点、依赖方、集成面

Agent提示模板：

Scope: [具体调研领域]
Objective: 挖掘与以下内容相关的[模式/约束/接口]：[需求摘要]
Output: 带有文件路径的研究结论列表
Context: 研究表明：[关键结论摘要]
Exclusions: 请勿分析实现细节或具体文件内容

[关键要求]:

[必须] 在同一条消息中调度所有Agent。
[必须] 仅聚焦于模式/约束/接口，而非具体实现。
[禁止] 深入研究文件内容——这是plan阶段的工作。

[3][EXPLORE]

Dictum: Comparison reveals optimal trade-offs.

Generate 2-3 distinct approaches from research + scan findings.

Per Approach:

[INDEX]	Aspect	Content
[1]	Strategy	High-level implementation direction
[2]	Alignment	How it leverages research findings
[3]	Patterns	Which codebase conventions it follows
[4]	Trade-offs	Pros and cons

Approach Generation Criteria:

Approach A: Most aligned with existing patterns (conservative)
Approach B: Best leverage of research findings (optimal)
Approach C: Simplest implementation path (minimal) — optional

[IMPORTANT]:

[ALWAYS] Ground approaches in scan findings—no speculation.
[ALWAYS] Include trade-off analysis per approach.
[ALWAYS] Apply YAGNI—cut unnecessary scope from all approaches.
[NEVER] Generate approaches without codebase evidence.

准则： 对比分析可揭示最优的取舍方案。

结合研究成果与扫描结果，生成2-3种不同的实现方案。

每个方案包含：

[序号]	维度	内容
[1]	策略	高层次实现方向
[2]	匹配度	如何利用研究成果
[3]	模式	遵循哪些代码库规范
[4]	优劣分析	优点与缺点

方案生成标准：

方案A：与现有模式最匹配（保守型）
方案B：最能充分利用研究成果（优化型）
方案C：实现路径最简单（极简型）——可选

[重要提示]:

[必须] 所有方案均基于扫描结果，不得凭空猜测。
[必须] 每个方案都包含优劣分析。
[必须] 应用YAGNI原则——移除所有方案中的非必要范围。
[禁止] 在没有代码库证据支持的情况下生成方案。

[4][SELECT]

Dictum: Committed direction enables focused planning.

Select best approach via weighted criteria:

[INDEX]	Criterion	Weight	Evaluation
[1]	Pattern alignment	High	Matches existing codebase conventions
[2]	Research support	High	Backed by high-confidence findings
[3]	Simplicity	Medium	Minimal moving parts
[4]	Risk profile	Medium	Low-confidence areas minimized

Selection Output:

Selected approach name
Primary rationale (1-2 sentences)
Key trade-off accepted

[CRITICAL]:

[ALWAYS] Commit to ONE approach—no hedging.
[ALWAYS] Document trade-off accepted.
[NEVER] Defer selection to downstream phases.

准则： 明确的方向可助力聚焦规划。

通过加权标准选择最优方案：

[序号]	评估标准	权重	评估方式
[1]	模式匹配度	高	是否符合现有代码库规范
[2]	研究支持度	高	是否有高置信度研究结论支撑
[3]	简洁性	中	移动部件最少
[4]	风险程度	中	低置信度领域的影响最小化

选择结果输出：

选中方案的名称
核心理由（1-2句话）
接受的关键取舍

[关键要求]:

[必须] 仅选择一个方案，不得含糊其辞。
[必须] 记录所接受的取舍。
[禁止] 将选择推迟到后续阶段。

[5][OUTPUT]

Dictum: Downstream consumers require predictable structure.

Produce

brainstorm.md

with structure:

markdown

undefined

准则： 下游环节需要可预测的文档结构。

生成

brainstorm.md

，结构如下：

markdown

undefined

[H1][DESIGN]: [Title]

[H1][DESIGN]: [标题]

Dictum: [Build target—refined from request]

Research Summary: [Key findings relevant to design]

准则： [从需求提炼出的构建目标]

研究摘要： [与设计相关的关键结论]

[1][APPROACHES]

[1.1][APPROACH_A]: [Name]

[1.1][APPROACH_A]: [名称]

[INDEX]	[ASPECT]	[DETAIL]
[1]	Strategy	[High-level direction]
[2]	Alignment	[Research findings leveraged]
[3]	Patterns	[Codebase conventions followed]
[4]	Pros	[Benefits]
[5]	Cons	[Drawbacks]

[序号]	[维度]	[详情]
[1]	策略	[高层次方向]
[2]	匹配度	[利用的研究结论]
[3]	模式	[遵循的代码库规范]
[4]	优点	[优势]
[5]	缺点	[劣势]

[1.2][APPROACH_B]: [Name]

[1.2][APPROACH_B]: [名称]

[INDEX]	[ASPECT]	[DETAIL]
[1]	Strategy	[High-level direction]
[2]	Alignment	[Research findings leveraged]
[3]	Patterns	[Codebase conventions followed]
[4]	Pros	[Benefits]
[5]	Cons	[Drawbacks]

[序号]	[维度]	[详情]
[1]	策略	[高层次方向]
[2]	匹配度	[利用的研究结论]
[3]	模式	[遵循的代码库规范]
[4]	优点	[优势]
[5]	缺点	[劣势]

[2][SELECTED_APPROACH]

[INDEX]	[KEY]	[VALUE]
[1]	Choice	[Approach name]
[2]	Rationale	[Why this approach]
[3]	Trade-off Accepted	[What we're giving up]

[序号]	[关键项]	[内容]
[1]	选择方案	[方案名称]
[2]	选择理由	[选择该方案的原因]
[3]	接受的取舍	[我们放弃的内容]

[3][DESIGN_CONSTRAINTS]

[INDEX]	[CONSTRAINT]	[SOURCE]
[1]	[Hard boundary]	[Codebase scan]
[2]	...	...

[序号]	[约束条件]	[来源]
[1]	[硬性边界]	[代码库扫描]
[2]	...	...

[4][KEY_DECISIONS]

[INDEX]	[DECISION]	[CHOICE]	[RATIONALE]
[1]	[Design choice]	[Selected option]	[Why]
[2]	[Design choice]	[Selected option]	[Why]


[CRITICAL]:
- [ALWAYS] Include all sections—downstream depends on structure.
- [ALWAYS] Table format for approaches and decisions.
- [NEVER] Prose paragraphs—tables and lists only.

---

[序号]	[设计决策]	[选择]	[理由]
[1]	[设计选项]	[选中的选项]	[原因]
[2]	[设计选项]	[选中的选项]	[原因]


[关键要求]:
- [必须] 包含所有章节——下游环节依赖该结构。
- [必须] 方案和决策使用表格格式。
- [禁止] 使用段落式文字——仅使用表格和列表。

---

design-synthesis

Original

Translation

[H1][DESIGN-SYNTHESIS]

[H1][DESIGN-SYNTHESIS]

[1][INGEST]

[1][INGEST]

[2][SCAN]

[2][SCAN]

[3][EXPLORE]

[3][EXPLORE]

[4][SELECT]

[4][SELECT]

[5][OUTPUT]

[5][OUTPUT]

[H1][DESIGN]: [Title]

[H1][DESIGN]: [标题]

[1][APPROACHES]

[1][APPROACHES]

[1.1][APPROACH_A]: [Name]

[1.1][APPROACH_A]: [名称]

[1.2][APPROACH_B]: [Name]

[1.2][APPROACH_B]: [名称]

[2][SELECTED_APPROACH]

[2][SELECTED_APPROACH]

[3][DESIGN_CONSTRAINTS]

[3][DESIGN_CONSTRAINTS]

[4][KEY_DECISIONS]

[4][KEY_DECISIONS]

[6][VALIDATION]

[6][VALIDATION]