self-improve

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Self-improve skill

Self-improve 技能

Activated by

blazor-architect

when

--self-improve

is set. Runs after all specialist work and the review loop have completed.

当设置

--self-improve

参数时，由

blazor-architect

激活。在所有专项工作和审核循环完成后运行。

Step 1 — Create the run directory

步骤1 — 创建运行目录

Create

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

if it does not exist.

Done when: the run directory exists on disk.

如果

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

完成标志： 运行目录已在磁盘上存在。

Step 2 — Generate the improvement report

步骤2 — 生成改进报告

Consult

references/self-improve-generation.md

for the complete algorithm. The ordered steps:

Collect all
```
self_diagnosis.issues
```
entries from specialist reports in the run.
Map each to a raw finding (title, summary, category, severity, expected_impact, prompt_fragment, evidence).

Derive

suggestion_key

for each finding (

<category>:<target_surface>:<normalized_intent>

Load

~/.self-improve-reports/blazor-architect/suggestion-history.json

if present.

Hard-exclude findings with
```
never_again
```
history entries.
Group by
```
suggestion_key
```
, fold using merge rules (max severity, evidence union, recurrence count).
Apply
```
history_weight
```
from most recent decision per key.
Compute
```
ranking_score
```
(base_severity_weight + recurrence_boost + history_weight).
Sort by severity group, then ranking_score descending, then first_seen ascending.
Assign sequential ids (f-001, f-002, ...).
Build the origin block from current run context.
Write
```
improvement-report-data.json
```
to the run directory.

The file must conform to

references/improvement-report-data-schema.json

(schema version 1.1). A valid example is at

references/improvement-report-data-example.json

Initial file shape:

json

{
  "schema_version": "1.1",
  "generated_at": "<ISO timestamp>",
  "origin": {
    "skill_id": "<active skill id>",
    "skill_scope": "repo | user",
    "skill_path": "<absolute path to skill directory>",
    "repo_root": "<absolute git repository root>",
    "run_id": "<run_id>"
  },
  "findings": [ ... ],
  "decisions": {},
  "shipped_prompt": null
}

decisions

and

shipped_prompt

are always empty/null at generation time — the C# server writes them after user actions.

Done when:

improvement-report-data.json

is written to the run directory with valid findings (or an empty findings array if no self-diagnosis issues were collected).

完整算法请参考

references/self-improve-generation.md

。有序步骤如下：

收集本次运行中所有专项报告里的
```
self_diagnosis.issues
```
条目。
将每个条目映射为原始发现项（包含标题、摘要、分类、严重程度、预期影响、提示片段、证据）。

为每个发现项生成

suggestion_key

（格式为

<category>:<target_surface>:<normalized_intent>

）。

如果存在

~/.self-improve-reports/blazor-architect/suggestion-history.json

文件，则加载该文件。

硬排除包含
```
never_again
```
历史记录的发现项。
按
```
suggestion_key
```
分组，使用合并规则合并（取最高严重程度、合并证据、统计重复次数）。
应用每个键对应最近决策的
```
history_weight
```
。
计算
```
ranking_score
```
（基础严重程度权重 + 重复次数增益 + 历史权重）。
先按严重程度分组排序，再按
```
ranking_score
```
降序排序，最后按首次出现时间升序排序。
分配连续ID（f-001、f-002……）。
根据当前运行上下文构建来源块。
将
```
improvement-report-data.json
```
写入运行目录。

该文件必须符合

references/improvement-report-data-schema.json

（schema版本1.1）。有效示例可查看

references/improvement-report-data-example.json

。

初始文件结构：

json

{
  "schema_version": "1.1",
  "generated_at": "<ISO timestamp>",
  "origin": {
    "skill_id": "<active skill id>",
    "skill_scope": "repo | user",
    "skill_path": "<absolute path to skill directory>",
    "repo_root": "<absolute git repository root>",
    "run_id": "<run_id>"
  },
  "findings": [ ... ],
  "decisions": {},
  "shipped_prompt": null
}

decisions

和

shipped_prompt

在生成时始终为空/Null——用户操作后由C#服务器写入内容。

完成标志：

improvement-report-data.json

已写入运行目录，且包含有效发现项（如果未收集到任何自我诊断问题，则为空发现项数组）。

Step 3 — Launch the report-server

步骤3 — 启动报告服务器

The report-server binary lives at

~/.copilot/gman-skills/bin/report-server.exe

(Windows) or

~/.copilot/gman-skills/bin/report-server

(Linux/macOS). Launch it with the report file path:

Windows:

~/.copilot/gman-skills/bin/report-server.exe --report-path ~/.self-improve-reports/blazor-architect/runs/<run_id>/improvement-report-data.json

Linux / macOS:

~/.copilot/gman-skills/bin/report-server --report-path ~/.self-improve-reports/blazor-architect/runs/<run_id>/improvement-report-data.json

The server listens on

127.0.0.1:5173

by default. If the port is already bound, assume the server is already running — log a warning and continue (do not launch a second instance).

After launching, open the browser to

http://127.0.0.1:5173

so the user can see the report. Announce the URL clearly:

[blazor-architect] Self-improvement report ready at http://127.0.0.1:5173
  report file: ~/.self-improve-reports/blazor-architect/runs/<run_id>/improvement-report-data.json
  server PID: <pid>

The server stays alive until

GET /shutdown

is called, an idle timeout elapses after the browser connects, or the terminal session ends. API surface:

GET /api/report

GET /ping

POST /api/ship-prompt

POST /api/dismissals

GET /shutdown

If the binary is missing, print a warning telling the user to run

/setup-gman-skills

and continue without the server — the report file is still useful on its own.

Done when: the server is running and responding on port 5173, or the binary is missing and a warning has been printed.

报告服务器二进制文件位于

~/.copilot/gman-skills/bin/report-server.exe

（Windows系统）或

~/.copilot/gman-skills/bin/report-server

（Linux/macOS系统）。使用报告文件路径启动服务器：

Windows系统：

~/.copilot/gman-skills/bin/report-server.exe --report-path ~/.self-improve-reports/blazor-architect/runs/<run_id>/improvement-report-data.json

Linux / macOS系统：

~/.copilot/gman-skills/bin/report-server --report-path ~/.self-improve-reports/blazor-architect/runs/<run_id>/improvement-report-data.json

服务器默认监听

127.0.0.1:5173

。如果该端口已被占用，则假定服务器已在运行——记录警告并继续（不要启动第二个实例）。

启动后，打开浏览器访问

http://127.0.0.1:5173

，以便用户查看报告。清晰告知URL：

[blazor-architect] 自我改进报告已就绪，访问地址：http://127.0.0.1:5173
  报告文件：~/.self-improve-reports/blazor-architect/runs/<run_id>/improvement-report-data.json
  服务器PID：<pid>

服务器会保持运行，直到调用

GET /shutdown

、浏览器连接后触发空闲超时，或终端会话结束。API接口包括：

GET /api/report

、

GET /ping

、

POST /api/ship-prompt

、

POST /api/dismissals

、

GET /shutdown

。

如果二进制文件缺失，打印警告告知用户运行

/setup-gman-skills

，然后不启动服务器继续执行——报告文件本身仍有使用价值。

完成标志： 服务器已在5173端口运行并响应，或二进制文件缺失且已打印警告。

Step 4 — Staging readiness (after server session ends)

步骤4 — 暂存就绪检查（服务器会话结束后）

After the server shuts down (user calls

GET /shutdown

, idle timeout, or terminal end), check the report file for staging readiness:

Ready when
```
decisions
```
is non-empty OR
```
shipped_prompt
```
is non-null.

When ready, stage:

git add ~/.self-improve-reports/blazor-architect/runs/<run_id>/improvement-report-data.json

If the file is already git-tracked and has local modifications, present the conflict flow (continue / stash / discard) from

references/self-improve-generation.md § Conflict flow

. Default to "Continue" after 30 seconds of no response.

Done when: staging is complete or the file is not ready to stage.

服务器关闭后（用户调用

GET /shutdown

、触发空闲超时，或终端结束），检查报告文件是否具备暂存条件：

就绪条件：
```
decisions
```
非空或
```
shipped_prompt
```
非Null。

就绪时，执行暂存：

git add ~/.self-improve-reports/blazor-architect/runs/<run_id>/improvement-report-data.json

如果文件已被Git追踪且存在本地修改，执行

references/self-improve-generation.md § Conflict flow

中的冲突处理流程（继续/暂存/丢弃）。如果30秒内无响应，默认选择“继续”。

完成标志： 暂存已完成，或文件暂未达到就绪条件。

Suggestion history

建议历史记录

The cross-run suggestion history store lives at

~/.self-improve-reports/blazor-architect/suggestion-history.json

. It records user decisions (

accepted

dismissed

never_again

) per

suggestion_key

across runs. It is never committed to source control. See

references/suggestion-history-schema.json

for the schema.

跨运行的建议历史记录存储在

~/.self-improve-reports/blazor-architect/suggestion-history.json

。它记录了不同运行中每个

suggestion_key

对应的用户决策（

accepted

、

dismissed

、

never_again

）。该文件不会提交到源代码控制。schema详情请查看

references/suggestion-history-schema.json

。