sdd-context-intake

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Context intake for Spec-Driven Development

规范驱动开发（Spec-Driven Development）的上下文采集

The user rarely starts with a clean one-line goal — they have a ticket, a doc, a thread, some files. This skill turns that raw material into the structured brief the SDD flow needs, WITHOUT guessing past what the sources actually say.

用户很少会从清晰的一句目标描述开始——他们通常会提供工单、文档、对话线程或一些文件。本技能将这些原始素材转换为SDD流程所需的结构化简报，且不会超出原始素材内容进行猜测。

When to use

使用场景

The user pastes a ticket / requirements / design doc / notes and says "do this".
The user names files, paths, or URLs as the source of truth.
Before
```
sdd-clarify
```
and before launching
```
sdd-feature-flow
```
, when the goal is buried in material rather than stated.

用户粘贴工单/需求/设计文档/纪要并说“据此执行”时。
用户将文件、路径或URL作为事实来源时。
在启动
```
sdd-clarify
```
和
```
sdd-feature-flow
```
之前，当目标隐藏在素材中而非明确陈述时。

Procedure

操作流程

Collect every source the user gave, verbatim: pasted text, plus anything referenced —
```
Read
```
the named files,
```
WebFetch
```
the URLs, pull the ticket via its MCP/CLI if available. Do NOT stop at the pasted blurb if it points elsewhere.
Delegate the reading when it's large or multi-source. Spawn a read-only agent (the
```
Explore
```
or
```
general-purpose
```
agent) to ingest the sources and return a compact digest — this keeps the bulky material out of the main context. For a single short paste, read inline.
Synthesize a brief with exactly these sections, grounded ONLY in the sources (mark anything you inferred):
- Goal — one or two lines: what to build and why (the "what/why", not the "how").
- In scope / Out of scope — bullet lists drawn from the sources.
- Constraints — data model, contracts, integrations, naming, limits the sources state.
- Load-bearing unknowns — decisions the sources do NOT settle that would change code, contract, or template structure. These become the questions for
```
sdd-clarify
```
  .
- Sources — the files/URLs/tickets you actually read.
Hand off. Feed the unknowns to
```
sdd-clarify
```
(ask the user), then pass the resolved brief as the
```
goal
```
(and any confirmed params) to the
```
sdd-feature-flow
```
workflow.

收集用户提供的所有来源内容，一字不差：包括粘贴的文本，以及所有引用的内容——读取指定文件、通过
```
WebFetch
```
获取URL内容，若可行则通过MCP/CLI拉取工单信息。若粘贴内容指向其他资源，请勿仅停留在粘贴内容上。
当内容量大或来源多样时，委托读取任务。生成一个只读Agent（
```
Explore
```
或通用Agent）来处理这些来源并返回简洁摘要——这样可以避免大量素材占用主上下文。对于单个简短的粘贴内容，可直接在线读取。
生成包含以下确切部分的简报，内容完全基于来源素材（对任何推断内容进行标记）：
- 目标：1-2行文字：要构建什么以及原因（即“是什么/为什么”，而非“怎么做”）。
- 包含范围/排除范围：从来源素材中提取的项目符号列表。
- 约束条件：来源素材中说明的数据模型、契约、集成、命名规则和限制。
- 关键未知项：来源素材未明确的、会影响代码、契约或模板结构的决策。这些将成为
```
sdd-clarify
```
  的待解决问题。
- 来源：实际读取的文件/URL/工单。
交接工作。将未知项传递给
```
sdd-clarify
```
（询问用户），然后将已明确的简报作为
```
goal
```
（以及任何已确认的参数）传递给
```
sdd-feature-flow
```
工作流。

Rules

规则

Ground everything in the sources. If the source doesn't say it, it's an unknown, not an assumption — surface it, don't invent it. (This is the same discipline the harness's clarity gate enforces later; catching it here is cheaper.)
Never begin writing a spec or code from this skill — its only output is the brief.
Keep the digest tight; the point is to compress source material into decisions and unknowns.

所有内容均基于来源素材。若来源未提及，则视为未知项，而非假设——需明确指出，不得自行编造。（这与后续工具链的清晰度检查门所要求的准则一致；在此阶段发现问题成本更低。）
本技能绝不开始编写规格说明或代码——其唯一输出是简报。
摘要需简洁；核心是将来源素材压缩为已确定的决策和未知项。