blazor-architect

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Blazor architect skill

Blazor架构师技能

Step 1 — Announce the run

步骤1 — 启动运行

Generate a run ID (

run-YYYYMMDD-HHMM

, local time) and print it as the first output line:

[blazor-architect] run-20260802-1716 started

Done when: the run ID line is printed at the top of your response.

生成一个运行ID（格式为

run-YYYYMMDD-HHMM

，使用本地时间），并将其作为第一行输出：

[blazor-architect] run-20260802-1716 started

完成标志： 运行ID行显示在响应内容的最顶部。

Step 2 — Parse the work request and flags from the invocation text

步骤2 — 从调用文本中解析工作请求和标志

The user's task description is embedded in the invocation text — the same line they typed to fire the skill (e.g.

/blazor-architect implement a counter component in Components/Pages/Counter.razor --self-improve

). Extract two things:

The work request — the task description, stripped of the skill name and flags. In the example above, the work request is: "implement a counter component in Components/Pages/Counter.razor".

Flags — scanned from the invocation text:

Flag	Recognised	Default
Skip code review	`--skip-code-review` or "skip code review"	off (review runs)
Self-improvement	`--self-improve` or "self improve"	off

If no work request is found (the user typed just

/blazor-architect

with no task), ask what they want to do and stop — do not proceed to routing.

Done when: you have identified the work request text and set

skip_code_review

and

self_improve

booleans.

用户的任务描述嵌入在调用文本中——即用户触发此技能时输入的内容（例如

/blazor-architect implement a counter component in Components/Pages/Counter.razor --self-improve

）。需要提取两项内容：

工作请求 —— 去除技能名称和标志后的任务描述。在上述示例中，工作请求为："implement a counter component in Components/Pages/Counter.razor"。

标志 —— 从调用文本中扫描得到：

标志	识别形式	默认值
跳过代码评审	`--skip-code-review` 或 "skip code review"	关闭（执行评审）
自我优化	`--self-improve` 或 "self improve"	关闭

如果未找到工作请求（用户仅输入

/blazor-architect

而未指定任务），询问用户需求并终止流程——不要继续进行路由操作。

完成标志： 已识别工作请求文本，并设置

skip_code_review

和

self_improve

布尔值。

Step 3 — Route the work

步骤3 — 路由工作请求

Apply the route classifier in

references/routing-classifier.md

to the work request. The classifier picks a primary lane and decides inline vs delegate. Print a one-line routing summary so the user sees the decision:

Route: inline (component-author, 1 file)

Route: delegate (component-author → data-fetching-specialist, serial)

Done when: a lane is selected and the inline/delegate decision is made and announced.

将

references/routing-classifier.md

中的路由分类器应用于工作请求。分类器会选择主通道，并决定是内联执行还是委托执行。打印一行路由摘要，让用户看到决策结果：

Route: inline (component-author, 1 file)

或

Route: delegate (component-author → data-fetching-specialist, serial)

完成标志： 已选择通道，确定内联/委托决策并进行公告。

Step 4 — Execute the work

步骤4 — 执行工作

Inline execution

内联执行

For narrow tasks (single lane, ≤ 2 files, no broad repo discovery), do the work directly in this context. Use the dotnet-blazor plugin skills as guidance — read the relevant skill's SKILL.md for patterns and constraints before writing code.

对于范围较窄的任务（单通道、≤2个文件、无需广泛仓库检索），直接在此上下文内完成工作。以dotnet-blazor插件技能为指导——在编写代码前，阅读相关技能的SKILL.md文档，了解模式和约束。

Delegated execution

委托执行

For broader tasks, dispatch via the

task

tool. Each specialist lane receives a bounded task for one lane. Specialist lanes:

Lane	Scope	Guidance skill	Delegate?
component-author	Create a new component, parameters, lifecycle, CSS isolation	`dotnet-blazor:author-component`	Inline if ≤ 2 files
component-extractor	Extract sections from a page into reusable components	—	Always delegate
form-specialist	Forms, binding, validation, EditForm, @bind	`dotnet-blazor:collect-user-input`	Inline if 1–2 files
data-fetching-specialist	HttpClient, service abstractions, loading/error/empty states	`dotnet-blazor:fetch-and-send-data`	Always delegate

Each specialist must return a structured report matching

references/feedback-report-template.json

. On first validation failure, request one schema-repair retry; on second failure, mark the lane

failed_report_schema

and preserve raw output.

对于范围较广的任务，通过

task

工具进行分派。每个专业通道会接收一个限定范围的单通道任务。专业通道如下：

通道	范围	指导技能	是否可内联？
component-author	创建新组件、参数、生命周期、CSS隔离	`dotnet-blazor:author-component`	若≤2个文件则可内联
component-extractor	将页面中的部分内容提取为可复用组件	—	始终委托
form-specialist	表单、绑定、验证、EditForm、@bind	`dotnet-blazor:collect-user-input`	若1–2个文件则可内联
data-fetching-specialist	HttpClient、服务抽象、加载/错误/空状态	`dotnet-blazor:fetch-and-send-data`	始终委托

每个专业人员必须返回符合

references/feedback-report-template.json

格式的结构化报告。首次验证失败时，请求一次模式修复重试；第二次失败时，标记该通道为

failed_report_schema

并保留原始输出。

Fluent UI

Fluent UI is a cross-lane constraint, not a lane. When Fluent components are in scope, invoke the

fluentui-blazor

skill as an overlay alongside the active specialist.

Done when: all specialist work is complete and reports are validated. Every specialist report conforms to the feedback-report schema.

Fluent UI是跨通道约束，而非独立通道。当涉及Fluent组件时，在调用当前专业技能的同时叠加调用

fluentui-blazor

技能。

完成标志： 所有专业工作已完成，且报告通过验证。每份专业报告均符合反馈报告模式。

Step 5 — Review gate

步骤5 — 评审关卡

Check the

skip_code_review

flag that was parsed in step 2. This is a gate — make the decision before dispatching any review sub-agent.

When
skip_code_review
is true: do not spawn a review sub-agent. Set

review_outcome

"skipped"

. Proceed to step 6.

When
skip_code_review
is false: run the review loop — a dedicated review sub-agent via the

task

tool with

agent_type: "code-review"

. The review loop:

Maximum 3 passes (initial + 2 fix-and-review cycles).
Stop early when a pass returns zero actionable findings.
Actionable = concrete defect, file-specific, not stylistic. See
```
references/review-loop-contract.md
```
.

The decision is made here, not during execution — the review sub-agent is only dispatched when the flag was not set, and only after all specialist work is complete.

Done when: review is complete or skipped. Either zero actionable findings remain, or unresolved findings are recorded for the run summary.

检查步骤2中解析得到的

skip_code_review

标志。这是一个关卡——在分派任何评审子代理前做出决策。

当
skip_code_review
为true时：不生成评审子代理。将

review_outcome

设置为

"skipped"

。继续执行步骤6。

当
skip_code_review
为false时：运行评审循环——通过

task

工具调用专用评审子代理，设置

agent_type: "code-review"

。评审循环规则：

最多3轮（初始评审 + 2次修复-评审循环）。
若某一轮评审未发现可操作问题，则提前终止。
可操作问题指具体缺陷、针对特定文件的问题，而非风格问题。详见
```
references/review-loop-contract.md
```
。

决策在此阶段做出，而非执行阶段——仅当未设置跳过标志，且所有专业工作完成后，才会分派评审子代理。

完成标志： 评审已完成或已跳过。要么无剩余可操作问题，要么已记录未解决问题用于运行总结。

Step 6 — Write run artifacts

步骤6 — 生成运行工件

Write

analysis.json

~/.self-improve-reports/blazor-architect/runs/<run_id>/

containing:

```
run_id
```
,
```
final_status
```
(precedence: failed > blocked > partial > success)
```
status_counts
```
,
```
lane_outcomes[]
```
```
review_outcome
```
(passed / fixed / review_unresolved / skipped)
```
follow_up_items[]
```
when status is partial or blocked

Create the run directory if it does not exist.

Done when:

analysis.json

is written to the run directory.

将

analysis.json

写入

~/.self-improve-reports/blazor-architect/runs/<run_id>/

目录，文件应包含：

```
run_id
```
、
```
final_status
```
（优先级：failed > blocked > partial > success）
```
status_counts
```
、
```
lane_outcomes[]
```
```
review_outcome
```
（passed / fixed / review_unresolved / skipped）
当状态为partial或blocked时，包含
```
follow_up_items[]
```

若运行目录不存在则创建。

完成标志：

analysis.json

已写入运行目录。

Step 7 — Self-improvement (only when

--self-improve

is set)

步骤7 — 自我优化（仅当设置

--self-improve

时执行）

When

--self-improve

is active, load the

self-improve

skill and execute its steps. The self-improve skill generates the improvement report and launches the report-server. Follow self-improve's steps exactly — do not inline its algorithm here.

When

--self-improve

is not set, skip this step entirely.

Done when: self-improve steps are complete (report generated, server launched) OR the flag was not set.

当

--self-improve

激活时，加载

self-improve

技能并执行其步骤。自我优化技能会生成改进报告并启动报告服务器。严格遵循自我优化技能的步骤——不要在此处内联其算法。

当未设置

--self-improve

时，完全跳过此步骤。

完成标志： 自我优化步骤已完成（报告生成、服务器启动），或未设置该标志。

Step 8 — Print the run summary

步骤8 — 打印运行总结

Close every run with a summary block. This is the user's primary feedback — make it informative:

[blazor-architect] run-20260802-1716 complete
  lanes:     component-author
  status:    success
  review:    passed
  files:     src/Konqvist.Web/Components/Pages/Counter.razor (created)
  artifacts: ~/.self-improve-reports/blazor-architect/runs/run-20260802-1716/analysis.json
  follow-up: none

When

--self-improve

was active, also include:

  report:    ~/.self-improve-reports/blazor-architect/runs/run-20260802-1716/improvement-report-data.json
  server:    http://127.0.0.1:5173/api/report (running)

When

final_status

partial

blocked

, surface blocked/failed lane names and recommended next actions.

Done when: the summary block is printed with all fields populated.

每个运行结束时输出总结块。这是用户的主要反馈信息——需确保内容详实：

[blazor-architect] run-20260802-1716 complete
  lanes:     component-author
  status:    success
  review:    passed
  files:     src/Konqvist.Web/Components/Pages/Counter.razor (created)
  artifacts: ~/.self-improve-reports/blazor-architect/runs/run-20260802-1716/analysis.json
  follow-up: none

当

--self-improve

激活时，还需添加：

  report:    ~/.self-improve-reports/blazor-architect/runs/run-20260802-1716/improvement-report-data.json
  server:    http://127.0.0.1:5173/api/report (running)

当

final_status

为

partial

或

blocked

时，需显示受阻/失败的通道名称以及建议的后续操作。

完成标志： 已打印填充所有字段的总结块。",

blazor-architect

Original

Translation

Blazor architect skill

Blazor架构师技能

Step 1 — Announce the run

步骤1 — 启动运行

Step 2 — Parse the work request and flags from the invocation text

步骤2 — 从调用文本中解析工作请求和标志

Step 3 — Route the work

步骤3 — 路由工作请求

Step 4 — Execute the work

步骤4 — 执行工作

Inline execution

内联执行

Delegated execution

委托执行

Fluent UI

Fluent UI

Step 5 — Review gate

步骤5 — 评审关卡

Step 6 — Write run artifacts

步骤6 — 生成运行工件

Step 7 — Self-improvement (only when
`--self-improve`
is set)

步骤7 — 自我优化（仅当设置
`--self-improve`
时执行）

Step 8 — Print the run summary

步骤8 — 打印运行总结

blazor-architect

Original

Translation

Blazor architect skill

Blazor架构师技能

Step 1 — Announce the run

步骤1 — 启动运行

Step 2 — Parse the work request and flags from the invocation text

步骤2 — 从调用文本中解析工作请求和标志

Step 3 — Route the work

步骤3 — 路由工作请求

Step 4 — Execute the work

步骤4 — 执行工作

Inline execution

内联执行

Delegated execution

委托执行

Fluent UI

Fluent UI

Step 5 — Review gate

步骤5 — 评审关卡

Step 6 — Write run artifacts

步骤6 — 生成运行工件

Step 7 — Self-improvement (only when --self-improve is set)

步骤7 — 自我优化（仅当设置--self-improve时执行）

Step 8 — Print the run summary

步骤8 — 打印运行总结

Step 7 — Self-improvement (only when
`--self-improve`
is set)

步骤7 — 自我优化（仅当设置
`--self-improve`
时执行）