gsdl

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

GSD — Get Shit Done Orchestrator

GSD — 高效落地编排器

Coordinates the full project pipeline across five phases. Phases 0–2 and Phase 4 run inline (interactive, in this context). Phase 3 (implementation) spawns one subagent per parent task to keep context fresh.

可协调覆盖5个阶段的完整项目管线，阶段0-2和阶段4内联执行（在当前上下文下交互运行），阶段3（功能实现）会为每个父任务创建独立子Agent以保证上下文新鲜准确。

Pipeline Overview

管线概览

Phase 0: Setup       → gsdl-setup-project SKILL.md        (inline)
Phase 1: PRD         → gsdl-create-prd SKILL.md           (inline, interactive Q&A)
Phase 2: Task List   → gsdl-create-plan SKILL.md          (inline, two-phase with "Go")
Phase 3: Implement   → gsdl-execute-plan SKILL.md         (subagent per parent task)
Phase 4: Document    → gsdl-document-decisions SKILL.md   (inline, optional publish to Slite/Notion)

Phase 0: 项目初始化       → gsdl-setup-project SKILL.md        (内联执行)
Phase 1: PRD输出          → gsdl-create-prd SKILL.md           (内联执行，交互式问答)
Phase 2: 任务清单生成      → gsdl-create-plan SKILL.md          (内联执行，用户确认"Go"后进入下一环节)
Phase 3: 功能实现         → gsdl-execute-plan SKILL.md         (每个父任务对应一个子Agent)
Phase 4: 文档归档         → gsdl-document-decisions SKILL.md   (内联执行，支持可选发布到Slite/Notion)

Step 0: Detect Source URL (Pre-flight)

步骤0：检测源URL（启动前检查）

Before resolving the project name, check if the user's message contains a URL argument.

Recognized patterns:

```
/gsdl linear https://linear.app/...
```
```
/gsdl notion https://notion.so/...
```
```
/gsdl slite https://slite.com/...
```
```
/gsdl https://linear.app/...
```
(source auto-detected from URL)
```
/gsdl https://notion.so/...
```
```
/gsdl https://notion.site/...
```
```
/gsdl https://slite.com/...
```

If a URL is detected:

Read and follow the
```
gsdl-fetch-source
```
skill (resolve path per the Skill Path Resolution section below).
Pass the URL and optional source-type hint to the skill.
The skill returns:
- A suggested project name (kebab-case, derived from the fetched title)
- Seed content (pre-formatted Markdown for
```
seed.md
```
  )
- A source summary (one-line description of what was fetched)
Present the suggested project name to the user and confirm: "I fetched the content from [source summary]. I'll use
[suggested-name]
as the project name — does that work, or would you like a different name?"
Once confirmed (or adjusted), store both the project name and seed content in context.
Proceed to Step 2: Detect Current Phase — the seed content will be written to
```
seed.md
```
during Phase 0 setup (overriding the normal "describe your idea" prompt).

If no URL is detected, skip this step entirely and proceed to Step 1.

在确定项目名之前，先检查用户消息中是否包含URL参数。

可识别的格式：

```
/gsdl linear https://linear.app/...
```
```
/gsdl notion https://notion.so/...
```
```
/gsdl slite https://slite.com/...
```
```
/gsdl https://linear.app/...
```
(自动从URL识别来源)
```
/gsdl https://notion.so/...
```
```
/gsdl https://notion.site/...
```
```
/gsdl https://slite.com/...
```

如果检测到URL：

读取并遵循
```
gsdl-fetch-source
```
技能的要求（按下方技能路径解析规则查找对应文件）。
将URL和可选的来源类型提示传入该技能。
技能会返回：
- 推荐项目名（kebab-case格式，从抓取的标题派生）
- 种子内容（预格式化的Markdown内容，用于写入
```
seed.md
```
  ）
- 来源摘要（抓取内容的一行描述）
向用户展示推荐的项目名并确认："我已经从[来源摘要]抓取了对应内容，将使用
[推荐名称]
作为项目名，是否可行？或者你可以提供其他名称。"
获得确认（或调整为用户提供的名称）后，将项目名和种子内容存储到上下文中。
进入步骤2：检测当前阶段——种子内容会在阶段0初始化时写入
```
seed.md
```
（替换默认的「描述你的想法」提示）。

如果未检测到URL，直接跳过本步骤进入步骤1。

Step 1: Resolve Project Name

步骤1：确定项目名

Check the user's message for a project name argument. If the user typed
```
/gsdl my-project
```
, the project name is
```
my-project
```
. Use it directly — do not ask.
If a source URL was provided in Step 0, the project name was already resolved there — use it.
If no project name was provided and no URL was given, ask: "What's the project name?"
Normalize to kebab-case.

检查用户消息中的项目名参数，如果用户输入
```
/gsdl my-project
```
，则项目名直接为
```
my-project
```
，无需额外询问。
如果步骤0中提供了源URL，项目名已经在步骤0中确定，直接使用即可。
如果未提供项目名也没有URL，询问用户："请问项目名是什么？"
将项目名统一格式化为kebab-case。

Step 2: Detect Current Phase

步骤2：检测当前阶段

Read

.planning/[project-name]/progress.md

if it exists — this is the fastest way to resume.

progress.md

exists, extract:

```
phase
```
— last recorded phase (0–4 or "done")
```
prd
```
— path to the active PRD file
```
tasks
```
— path to the active task file

Then verify the recorded files still exist on disk before trusting them. If a file is missing, fall back to the filesystem scan below.

progress.md

does not exist (or a recorded file is missing), scan the filesystem:

Condition	Start at
No `.planning/[project-name]/` folder	Phase 0
`.planning/[project-name]/seed.md` exists, no `prd-*.md`	Phase 1
`.planning/[project-name]/prd-.md` exists, no `tasks-prd-.md`	Phase 2
`.planning/[project-name]/tasks-prd-*.md` exists with unchecked `[ ]` items	Phase 3
All tasks `[x]` , no `decisions-*.md` file	Phase 4
`decisions-[project-name].md` exists	Done

If multiple PRD or task files exist, use the path recorded in

progress.md

(if available); otherwise ask the user which one to use.

After determining the active phase and files, write (or update)
progress.md
— see the Progress File section below.

Announce which phase you're starting from and why before proceeding.

如果

.planning/[项目名]/progress.md

存在则优先读取该文件——这是恢复项目进度最快的方式。

如果

progress.md

存在，提取以下信息：

```
phase
```
— 上次记录的阶段（0-4或"done"）
```
prd
```
— 生效的PRD文件路径
```
tasks
```
— 生效的任务文件路径

在使用这些信息前验证记录的文件是否仍存在于磁盘中，如果文件丢失，按下方文件系统扫描规则降级处理。

如果

progress.md

不存在（或记录的文件丢失），扫描文件系统判断阶段：

判定条件	起始阶段
不存在 `.planning/[项目名]/` 文件夹	阶段0
存在 `.planning/[项目名]/seed.md` ，无 `prd-*.md` 文件	阶段1
存在 `.planning/[项目名]/prd-.md` ，无 `tasks-prd-.md` 文件	阶段2
存在 `.planning/[项目名]/tasks-prd-*.md` ，且存在未勾选的 `[ ]` 任务项	阶段3
所有任务已勾选 `[x]` ，无 `decisions-*.md` 文件	阶段4
存在 `decisions-[项目名].md` 文件	已完成

如果存在多个PRD或任务文件，优先使用

progress.md

中记录的路径（如果存在）；否则询问用户要使用哪个文件。

确定当前生效的阶段和文件后，写入（或更新）
progress.md
——参考下方进度文件章节的规则。

在继续执行前先告知用户当前从哪个阶段开始，以及对应的判定原因。

Step 3: Run Each Phase

步骤3：逐阶段执行

Skill Path Resolution

技能路径解析规则

When reading a sub-skill, resolve its path using this priority order:

```
~/.agents/skills/[skill-name]/SKILL.md
```
— installed via npx (preferred)
```
~/.cursor/skills/[skill-name]/SKILL.md
```
— legacy/manual install

Use whichever path exists. If both exist, prefer

~/.agents/skills/

Sub-skills used by this orchestrator:

```
gsdl-fetch-source
```
— fetches content from Linear, Notion, or Slite (Step 0)
```
gsdl-setup-project
```
— creates folder structure and seed.md (Phase 0)
```
gsdl-create-prd
```
— generates the PRD (Phase 1)
```
gsdl-create-plan
```
— generates the task list (Phase 2)
```
gsdl-execute-plan
```
— implements sub-tasks (Phase 3)
```
gsdl-document-decisions
```
— captures decisions & architecture changes, optionally publishes to Slite or Notion (Phase 4)

读取子技能时，按以下优先级解析路径：

```
~/.agents/skills/[技能名]/SKILL.md
```
— 通过npx安装的路径（优先）
```
~/.cursor/skills/[技能名]/SKILL.md
```
— 旧版本/手动安装路径

使用存在的路径即可，如果两个路径都存在，优先使用

~/.agents/skills/

下的文件。

本编排器使用的子技能：

```
gsdl-fetch-source
```
— 从Linear、Notion或Slite抓取内容（步骤0使用）
```
gsdl-setup-project
```
— 创建文件夹结构和seed.md（阶段0使用）
```
gsdl-create-prd
```
— 生成PRD（阶段1使用）
```
gsdl-create-plan
```
— 生成任务清单（阶段2使用）
```
gsdl-execute-plan
```
— 实现子任务（阶段3使用）
```
gsdl-document-decisions
```
— 记录决策和架构变更，支持可选发布到Slite或Notion（阶段4使用）

Phase 0 — Project Setup (Inline)

阶段0 — 项目初始化（内联执行）

Read and follow the

gsdl-setup-project

skill (resolve path per above).

Complete the full setup (folder,

seed.md

If seed content was pre-fetched in Step 0: Write the fetched seed content directly to
```
seed.md
```
instead of asking the user to describe their idea. Show the user what was written and let them know they can edit
```
seed.md
```
before continuing.
If no pre-fetched content: Follow the normal
```
gsdl-setup-project
```
flow (prompt the user to describe their idea, populate
```
seed.md
```
collaboratively).

Then write the initial

progress.md

(phase 0), show the checkpoint, and wait for user confirmation before continuing to Phase 1.

读取并遵循

gsdl-setup-project

技能的要求（按上述规则解析路径）。

完成完整初始化流程（创建文件夹、生成

seed.md

）：

如果步骤0中已经预抓取了种子内容：直接将抓取的种子内容写入
```
seed.md
```
，无需询问用户描述想法，向用户展示写入的内容并告知可以在继续前编辑
```
seed.md
```
。
如果没有预抓取内容：遵循标准的
```
gsdl-setup-project
```
流程（提示用户描述想法，协作填充
```
seed.md
```
）。

然后写入初始的

progress.md

（阶段0），展示检查点，等待用户确认后再进入阶段1。

Phase 1 — Create PRD (Inline)

阶段1 — 生成PRD（内联执行）

Read and follow the

gsdl-create-prd

skill (resolve path per above).

This phase is interactive — ask clarifying questions, iterate on the PRD with the user, then save it to

.planning/[project-name]/prd-[feature-name].md

After saving the PRD, update

progress.md

(phase 1, prd path). Show the checkpoint and wait for confirmation before Phase 2. Before proceeding to Phase 2, re-read the PRD from disk at its saved path — the user may have edited the file after reviewing it.

读取并遵循

gsdl-create-prd

技能的要求（按上述规则解析路径）。

本阶段为交互式：询问澄清问题，和用户迭代优化PRD，然后保存到

.planning/[项目名]/prd-[功能名].md

。

保存PRD后，更新

progress.md

（阶段1，记录PRD路径），展示检查点，等待用户确认后再进入阶段2。进入阶段2前，从磁盘重新读取保存的PRD文件——用户可能在审阅后编辑了文件。

Phase 2 — Generate Task List (Inline)

阶段2 — 生成任务清单（内联执行）

Read and follow the

gsdl-create-plan

skill (resolve path per above).

This has two steps: show parent tasks → wait for user "Go" → generate sub-tasks and save to

.planning/[project-name]/tasks-prd-[name].md

After saving the task file, update

progress.md

(phase 2, tasks path). Show the checkpoint and wait for confirmation before Phase 3. Before proceeding to Phase 3, re-read the task file from disk at its saved path — the user may have edited the file (reordering, rewording, or adding tasks) after reviewing it.

读取并遵循

gsdl-create-plan

技能的要求（按上述规则解析路径）。

本阶段分为两步：展示父任务 → 等待用户回复"Go"确认 → 生成子任务并保存到

.planning/[项目名]/tasks-prd-[名称].md

。

保存任务文件后，更新

progress.md

（阶段2，记录任务文件路径），展示检查点，等待用户确认后再进入阶段3。进入阶段3前，从磁盘重新读取保存的任务文件——用户可能在审阅后编辑了文件（调整顺序、修改描述、新增任务等）。

Phase 3 — Implementation (Subagent per Parent Task)

阶段3 — 功能实现（每个父任务对应一个子Agent）

Update

progress.md

to phase 3 before starting.

Re-read the task file from disk to identify all parent tasks (1.0, 2.0, 3.0, ...). Use the on-disk version as the source of truth, not any prior in-context copy.

For each parent task, in order:

Read the task file to get the parent task title and its sub-tasks
Spawn a
```
generalPurpose
```
subagent using the Task tool with this prompt (fill in all
```
[placeholders]
```
):

Read and follow the skill at: ~/.agents/skills/gsdl-execute-plan/SKILL.md (if it exists, otherwise ~/.cursor/skills/gsdl-execute-plan/SKILL.md)

You are running in BATCH MODE. Complete ALL sub-tasks under the assigned parent task
without pausing for user approval between sub-tasks. Apply the full completion protocol
after each sub-task (mark [x], update task file, update Relevant Files), but immediately
continue to the next sub-task without waiting.

Project context:
- Project name: [project-name]
- Workspace root: [absolute path to workspace root]
- Task file: [absolute path to .planning/[project-name]/tasks-prd-*.md]
- Assigned parent task: [N.0] [Parent Task Title]
- Sub-tasks to complete: [N.1] [title], [N.2] [title], ... (list all)

When all sub-tasks under [N.0] are marked [x] and the parent is marked [x]:
- Return a summary: what was built, files created/modified, any issues or blockers
- Do NOT start the next parent task ([N+1].0)

After the subagent returns, show the user the summary
Show the checkpoint and wait for confirmation before spawning the next subagent

开始前先将

progress.md

更新为阶段3。

从磁盘重新读取任务文件识别所有父任务（1.0、2.0、3.0...），以磁盘上的版本为准，不要使用上下文中的旧副本。

按顺序处理每个父任务：

读取任务文件获取父任务标题和对应的子任务
使用任务工具创建
```
generalPurpose
```
子Agent，传入以下提示（填充所有
```
[占位符]
```
）：

读取并遵循该路径下的技能要求：~/.agents/skills/gsdl-execute-plan/SKILL.md（如果不存在则使用~/.cursor/skills/gsdl-execute-plan/SKILL.md）

你当前运行在批处理模式下，完成分配的父任务下的所有子任务，子任务之间不需要暂停等待用户确认。每个子任务完成后执行完整的完成协议（标记为[x]、更新任务文件、更新相关文件），然后立即继续下一个子任务，无需等待。

项目上下文：
- 项目名：[项目名]
- 工作区根目录：[工作区根目录绝对路径]
- 任务文件：[.planning/[项目名]/tasks-prd-*.md的绝对路径]
- 分配的父任务：[N.0] [父任务标题]
- 待完成子任务：[N.1] [标题], [N.2] [标题], ... （列出所有子任务）

当[N.0]下的所有子任务都标记为[x]且父任务也标记为[x]后：
- 返回总结：实现的内容、创建/修改的文件、遇到的问题或阻塞点
- 不要开始下一个父任务（[N+1].0）

子Agent返回结果后，向用户展示执行总结
展示检查点，等待用户确认后再创建下一个子Agent

Phase 4 — Document Decisions (Inline)

阶段4 — 决策文档归档（内联执行）

After all parent tasks are complete, update

progress.md

to phase

, then run this phase.

Read and follow the

gsdl-document-decisions

skill (resolve path per above).

Pass the following context to the skill:

```
PROJECT_NAME
```
: the current project name
```
WORKSPACE_ROOT
```
: the absolute path to the workspace root
```
PRD_PATH
```
: the PRD path recorded in
```
progress.md
```
```
TASKS_PATH
```
: the tasks path recorded in
```
progress.md
```

The skill will:

Analyze git history and the PRD/task files

Generate a structured decisions document at

.planning/[project-name]/decisions-[project-name].md

Ask the user if they want to publish it to Slite or Notion
Handle publishing if the user provides a parent page URL

This phase runs inline (no subagent). After the decisions document is saved and the user has answered the publish prompt, update

progress.md

done

, then show the final checkpoint:

🎉 Project '[project-name]' is complete.

Decisions document: .planning/[project-name]/decisions-[project-name].md
[If published: Published to [Slite/Notion]: [URL]]

Run /gsdl [project-name] again if you want to resume or extend this project.

所有父任务完成后，将

progress.md

更新为阶段

，然后运行本阶段。

读取并遵循

gsdl-document-decisions

技能的要求（按上述规则解析路径）。

向技能传入以下上下文：

```
PROJECT_NAME
```
：当前项目名
```
WORKSPACE_ROOT
```
：工作区根目录绝对路径
```
PRD_PATH
```
：
```
progress.md
```
中记录的PRD路径
```
TASKS_PATH
```
：
```
progress.md
```
中记录的任务文件路径

该技能会：

分析git历史、PRD和任务文件

在

.planning/[项目名]/decisions-[项目名].md

生成结构化的决策文档

询问用户是否需要发布到Slite或Notion
如果用户提供了父页面URL则执行发布操作

本阶段内联执行（不需要子Agent）。决策文档保存完成且用户回复发布选项后，将

progress.md

更新为

done

，然后展示最终检查点：

🎉 项目'[项目名]'已完成。

决策文档路径：.planning/[项目名]/decisions-[项目名].md
[如果已发布：已发布到[Slite/Notion]：[URL]]

如果你想要恢复或扩展本项目，再次运行/gsdl [项目名]即可。

Progress File

进度文件

Maintain

.planning/[project-name]/progress.md

throughout the pipeline. Write it after every phase transition and whenever the active files are first determined.

整个管线运行过程中维护

.planning/[项目名]/progress.md

文件，每次阶段切换、首次确定生效文件时都要更新该文件。

Format

格式

markdown

undefined

markdown

undefined

[project-name] — GSDL Progress

[项目名] — GSDL 进度

State

状态

Phase: [0 | 1 | 2 | 3 | 4 | done]
Last Updated: [YYYY-MM-DD]

阶段：[0 | 1 | 2 | 3 | 4 | done]
上次更新：[YYYY-MM-DD]

Active Files

生效文件

PRD: [relative path to active prd-*.md, or "none"]
Tasks: [relative path to active tasks-prd-*.md, or "none"]

undefined

PRD：[生效prd-*.md的相对路径，或"none"]
任务文件：[生效tasks-prd-*.md的相对路径，或"none"]

undefined

When to write/update

更新时机

Event	Phase value	PRD	Tasks
Phase 0 complete (setup done)	`1`	`none`	`none`
PRD saved (Phase 1 complete)	`2`	path to PRD	`none`
Task file saved (Phase 2 complete)	`3`	path to PRD	path to tasks
All tasks complete (Phase 3 done)	`4`	path to PRD	path to tasks
Decisions document saved (Phase 4 done)	`done`	path to PRD	path to tasks

事件	阶段值	PRD	任务文件
阶段0完成（初始化完成）	`1`	`none`	`none`
PRD保存完成（阶段1完成）	`2`	PRD路径	`none`
任务文件保存完成（阶段2完成）	`3`	PRD路径	任务文件路径
所有任务完成（阶段3完成）	`4`	PRD路径	任务文件路径
决策文档保存完成（阶段4完成）	`done`	PRD路径	任务文件路径

Checkpoint Format

检查点格式

Show this between every phase transition and between every parent task in Phase 3:

✅ [Phase name / Task N.0 title] complete.

What was done: [1–3 sentence summary]
Files touched: [key files created or modified]

👉 Review the output file before continuing — you can edit it directly and your changes will be picked up automatically.

Ready to continue to [next phase / Task N+1.0]?
Type "yes" / "go" / "next" to continue, or tell me what to review or change first.

Never proceed past a checkpoint without user confirmation. Always re-read the relevant artifact from disk after the user confirms, to pick up any edits they made during the review window.

每次阶段切换之间、阶段3的每个父任务完成后都展示以下检查点：

✅ [阶段名 / 任务N.0标题] 已完成。

完成内容：[1-3句话总结]
涉及文件：[创建或修改的核心文件]

👉 继续前请审阅输出文件，你可以直接编辑文件，修改会自动生效。

是否准备好进入[下一阶段 / 任务N+1.0]？
输入"yes" / "go" / "next"继续，或者告知我需要先审阅或修改的内容。

没有用户确认绝对不要跳过检查点继续执行。用户确认后一定要从磁盘重新读取相关产物，以拉取用户审阅期间做的所有修改。

Resuming Mid-Pipeline

管线中途恢复

If GSD is triggered on a project that already has work done:

Read
```
progress.md
```
(if present) → verify files on disk → fall back to filesystem scan if needed
Skip completed phases
Announce: "Project '[name]' is at Phase [N]. Resuming from [description]."
For Phase 3, check which parent tasks are already
```
[x]
```
and skip them
For Phase 4, if
```
decisions-[project-name].md
```
already exists, report the project is done — do not re-run Phase 4

如果GSD被触发时项目已经有完成的内容：

读取
```
progress.md
```
（如果存在）→ 验证磁盘上的文件 → 必要时降级为文件系统扫描判断阶段
跳过已完成的阶段
告知用户："项目'[名称]'当前处于阶段[N]，从[描述]处恢复执行。"
对于阶段3，检查哪些父任务已经标记为
```
[x]
```
，直接跳过这些任务
对于阶段4，如果
```
decisions-[项目名].md
```
已经存在，告知用户项目已完成——不要重新运行阶段4

Rules

规则

Always show a checkpoint between phases and between parent tasks — never chain them automatically
Phases 0–2 and Phase 4 are inline — do not spawn subagents for setup, PRD, task generation, or documentation
Each Phase 3 subagent handles exactly one parent task — never assign two parent tasks to one subagent
Surface blockers immediately — if a subagent reports an error or missing dependency, pause Phase 3 and resolve with the user before continuing
All files stay within the project folder — never create files at workspace root
Keep
progress.md
current — update it at every phase transition so
```
/gsdl [project-name]
```
always resumes cleanly
Phase 4 is mandatory but publishing is optional — always generate the decisions document; only push to Slite/Notion if the user provides a parent URL

阶段之间、父任务之间必须展示检查点——绝对不要自动串联执行
阶段0-2和阶段4为内联执行——不要为初始化、PRD生成、任务清单生成、文档归档创建子Agent
阶段3的每个子Agent仅处理一个父任务——绝对不要给单个子Agent分配两个父任务
阻塞点立即暴露——如果子Agent报告错误或缺失依赖，暂停阶段3执行，和用户解决问题后再继续
所有文件都保存在项目文件夹内——绝对不要在工作区根目录创建文件
保持
progress.md
实时更新——每次阶段切换都要更新，保证
```
/gsdl [项目名]
```
总能干净恢复
阶段4是强制步骤但发布是可选的——必须生成决策文档；只有用户提供父URL时才推送到Slite/Notion