solo-pipeline

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

/pipeline

Launch an automated multi-skill pipeline. The Stop hook chains skills automatically — no manual invocation needed between stages.

启动自动化多Skill流水线。Stop钩子会自动串联各个Skill——各阶段之间无需手动调用。

Available Pipelines

可用流水线

Research Pipeline

研究流水线

/pipeline research "AI therapist app"

Chains:

/research

/validate

Produces:

research.md

prd.md

/pipeline research "AI therapist app"

串联流程：

/research

→

/validate

产出物：

research.md

→

prd.md

Dev Pipeline

开发流水线

/pipeline dev "project-name" "stack"

/pipeline dev "project-name" "stack" --feature "user onboarding"

Chains:

/scaffold

/setup

/plan

/build

Produces: full project with workflow, plan, and implementation

/pipeline dev "project-name" "stack"

/pipeline dev "project-name" "stack" --feature "user onboarding"

串联流程：

/scaffold

→

/setup

→

/plan

→

/build

产出物：包含工作流、规划和实现代码的完整项目

Steps

步骤

1. Parse Arguments

1. 解析参数

Extract from

$ARGUMENTS

Pipeline type: first word (
```
research
```
or
```
dev
```
)
Remaining args: passed to the launcher script

If no arguments or unclear, ask:

Which pipeline do you want to run?

1. Research Pipeline — /research → /validate (idea to PRD)
2. Dev Pipeline — /scaffold → /setup → /plan → /build (PRD to running code)

从

$ARGUMENTS

中提取：

流水线类型：第一个单词（
```
research
```
或
```
dev
```
）
剩余参数：传递给启动脚本

如果没有参数或参数不明确，询问用户：

你想要运行哪种流水线？

1. 研究流水线 — /research → /validate（从想法到PRD）
2. 开发流水线 — /scaffold → /setup → /plan → /build（从PRD到可运行代码）

2. Confirm with User

2. 与用户确认

Show what will happen:

Pipeline: {type}
Stages: {stage1} → {stage2} → ...
Idea/Project: {name}

This will run multiple skills automatically. Continue?

Ask via AskUserQuestion.

展示即将执行的操作：

流水线类型：{type}
阶段：{stage1} → {stage2} → ...
想法/项目：{name}

这将自动运行多个Skill，是否继续？

通过AskUserQuestion发起询问。

3. Run Launcher Script

3. 运行启动脚本

Determine the plugin root (where this skill lives):

Check if
```
${CLAUDE_PLUGIN_ROOT}
```
is set (plugin context)
Otherwise find
```
solo-factory/scripts/
```
relative to project

bash

undefined

确定插件根目录（本Skill所在位置）：

检查是否设置了
```
${CLAUDE_PLUGIN_ROOT}
```
（插件上下文）
否则查找项目相对路径下的
```
solo-factory/scripts/
```

bash

undefined

Research pipeline

研究流水线

${CLAUDE_PLUGIN_ROOT}/scripts/solo-research.sh "idea name" [--project name] --no-dashboard

Dev pipeline

开发流水线

${CLAUDE_PLUGIN_ROOT}/scripts/solo-dev.sh "project-name" "stack" [--feature "desc"] --no-dashboard


**Always pass `--no-dashboard`** when running from within Claude Code skill context (tmux is for terminal use only).

${CLAUDE_PLUGIN_ROOT}/scripts/solo-dev.sh "project-name" "stack" [--feature "desc"] --no-dashboard


**在Claude Code Skill上下文环境中运行时，务必传递`--no-dashboard`参数**（tmux仅适用于终端场景）。

4. Start First Stage

4. 启动第一阶段

After the script creates the state file, immediately run the first stage's skill. The Stop hook will handle subsequent stages automatically.

For research pipeline: Run

/research "idea name"

For dev pipeline: Run

/scaffold project-name stack

脚本创建状态文件后，立即运行第一阶段对应的Skill。后续阶段将由Stop钩子自动处理。

对于研究流水线：运行

/research "idea name"

对于开发流水线：运行

/scaffold project-name stack

5. Pipeline Completion

5. 流水线完成

When all stages are done, output:

<solo:done/>

The Stop hook checks for this signal and cleans up the state file.

所有阶段完成后，输出：

<solo:done/>

Stop钩子会检测该信号并清理状态文件。

State File

状态文件

Location:

.solo/pipelines/solo-pipeline-{project}.local.md

(project-local) or

~/.solo/pipelines/solo-pipeline-{project}.local.md

(global fallback) Log file:

.solo/pipelines/solo-pipeline-{project}.log

Format: YAML frontmatter with stages list,

project_root

, and

log_file

fields. The Stop hook reads this file on every session exit attempt.

To cancel a pipeline manually: delete the state file

solo-pipeline-{project}.local.md

存储位置：

.solo/pipelines/solo-pipeline-{project}.local.md

（项目本地）或

~/.solo/pipelines/solo-pipeline-{project}.local.md

（全局 fallback）日志文件：

.solo/pipelines/solo-pipeline-{project}.log

格式：包含阶段列表、

project_root

和

log_file

字段的YAML前置内容。Stop钩子会在每次会话退出尝试时读取该文件。

手动取消流水线：删除状态文件

solo-pipeline-{project}.local.md

Monitoring

监控

tmux Dashboard (terminal use)

Tmux 仪表盘（终端场景）

When launched from terminal (without

--no-dashboard

), a tmux dashboard opens automatically with:

Pane 0: work area
Pane 1:
```
tail -f
```
on log file
Pane 2: live status display (refreshes every 2s)

If solo-factory scripts are available, manual dashboard commands:

bash

undefined

在终端启动时（不带

--no-dashboard

参数），会自动打开tmux仪表盘，包含：

面板0：工作区
面板1：日志文件的
```
tail -f
```
实时输出
面板2：实时状态显示（每2秒刷新一次）

如果solo-factory脚本可用，可使用以下手动仪表盘命令：

bash

undefined

Create dashboard for a pipeline

为流水线创建仪表盘

solo-dashboard.sh create <project>

Attach to existing dashboard

连接到已存在的仪表盘

solo-dashboard.sh attach <project>

Close dashboard

关闭仪表盘

solo-dashboard.sh close <project>

undefined

solo-dashboard.sh close <project>

undefined

Manual Monitoring

手动监控

If solo-factory scripts are available:

bash

undefined

如果solo-factory脚本可用：

bash

undefined

Colored status display

彩色状态显示

solo-pipeline-status.sh # all pipelines solo-pipeline-status.sh <project> # specific pipeline

solo-pipeline-status.sh # 所有流水线 solo-pipeline-status.sh <project> # 指定流水线

Auto-refresh

自动刷新

watch -n2 -c solo-pipeline-status.sh


Otherwise, use standard tools:
```bash

watch -n2 -c solo-pipeline-status.sh


否则，使用标准工具：
```bash

Log tail

实时查看日志

tail -f .solo/pipelines/solo-pipeline-<project>.log

Check state file

查看状态文件

cat .solo/pipelines/solo-pipeline-<project>.local.md

undefined

cat .solo/pipelines/solo-pipeline-<project>.local.md

undefined

Real-time Tool Visibility

实时工具可见性

If solo-factory scripts are available, the pipeline uses

--output-format stream-json

piped through

solo-stream-fmt.py

— tool calls appear in real-time with colored icons:

  Read docs/research.md
  Glob "*.md" ~/projects/my-app/
  Bash npm test
  WebSearch voice AI agent developer tools 2026
  Task [Explore] Research task
  kb_search voice agent

Disable colors:

--no-color

. Disable sound effects:

--no-sound

如果solo-factory脚本可用，流水线会使用

--output-format stream-json

并通过

solo-stream-fmt.py

处理——工具调用会实时显示并带有彩色图标：

  Read docs/research.md
  Glob "*.md" ~/projects/my-app/
  Bash npm test
  WebSearch voice AI agent developer tools 2026
  Task [Explore] Research task
  kb_search voice agent

禁用颜色：

--no-color

。禁用音效：

--no-sound

。

Session Reuse

会话复用

Re-running a pipeline reuses the existing tmux session:

All panes are cleared (Ctrl-C + clear)
Log tail and status watch restart fresh
No need to close/recreate — just run the same command again

重新运行流水线会复用已有的tmux会话：

所有面板会被清空（Ctrl-C + clear）
日志实时查看和状态监控会重新启动
无需关闭/重新创建会话——只需再次运行相同命令即可

Log Format

日志格式

[22:30:15] START    | my-app | stages: research -> validate | max: 5
[22:30:16] STAGE    | iter 1/5 | stage 1/2: research
[22:30:16] INVOKE   | /research "AI therapist app"
[22:35:42] CHECK    | research | .../research.md -> FOUND
[22:35:42] STAGE    | iter 2/5 | stage 2/2: validate
[22:35:42] INVOKE   | /validate "AI therapist app"
[22:40:10] CHECK    | validate | .../prd.md -> FOUND
[22:40:10] DONE     | All stages complete! Promise detected.
[22:40:10] FINISH   | Duration: 10m

[22:30:15] START    | my-app | stages: research -> validate | max: 5
[22:30:16] STAGE    | iter 1/5 | stage 1/2: research
[22:30:16] INVOKE   | /research "AI therapist app"
[22:35:42] CHECK    | research | .../research.md -> FOUND
[22:35:42] STAGE    | iter 2/5 | stage 2/2: validate
[22:35:42] INVOKE   | /validate "AI therapist app"
[22:40:10] CHECK    | validate | .../prd.md -> FOUND
[22:40:10] DONE     | All stages complete! Promise detected.
[22:40:10] FINISH   | Duration: 10m

Critical Rules

关键规则

Always confirm before starting a pipeline.
Don't skip stages — the hook handles progression.
Cancel = delete state file — tell users this if they want to stop.
Max iterations prevent infinite loops (default 5 for research, 15 for dev).
Use
--no-dashboard
when running from within Claude Code skill context.

启动前务必确认：启动流水线前必须与用户确认。
请勿跳过阶段：阶段推进由钩子自动处理。
取消方式 = 删除状态文件：如果用户想要停止流水线，请告知此方法。
最大迭代次数：防止无限循环（研究流水线默认5次，开发流水线默认15次）。
在Claude Code Skill上下文环境中运行时，务必使用
--no-dashboard
参数。