新功能：用 Skills 把开发工作整理成可学习的英语表达

---
name: english-learning
description: |
  Extract 1-5 reusable workplace English expressions from the current conversation, ask the user which ones to keep, and upload the approved sentences to LangCapture. Use when the user asks to "提炼英语", "提取英语表达", "学习英语", "上传到 LangCapture", "整理英语句子", "收集英语素材", or similar requests about reviewing the current session and turning it into English learning material.
---

# English Learning

Extract concise, reusable workplace English from the current chat. Confirm with the user before uploading anything to LangCapture.

## Workflow

### 1. Review the current session

Read the conversation history in the current session. Identify the main work topics being discussed, such as debugging, implementation, status updates, review feedback, deployment, testing, or coordination.

Think from the perspective of explaining the same work to an English-speaking coworker. Prefer expressions that are likely to be reused in engineering conversations.

### 2. Extract 1-5 useful English expressions

Select only sentences that are:

- Natural workplace English
- Reusable without depending on hidden context
- Useful for engineers discussing progress, issues, requests, review, testing, or results
- Concise enough to remember and practice

Avoid sentences that are:

- Overly academic or textbook-like
- Too specific to internal file names, secrets, or one-off details
- Unclear without extra explanation

For each item, output:

1. The English sentence
2. A short Chinese note describing meaning or usage

Use this format:

~~~text
1. "The latest change fixes the retry path, but I still need to verify the error handling."
   -> 用于汇报修复进展，同时说明还有待验证的部分
~~~

If the session does not contain enough good material, say so directly instead of inventing weak examples.

### 3. Ask for confirmation before upload

After listing the extracted expressions, ask the user which ones to upload.

If an interactive question tool is available, use it with these options:

- All upload
- Choose specific items
- Skip

If no such tool is available, ask a short plain-text question that offers the same three choices.

Do not upload anything until the user confirms.

If the user skips, stop immediately.

### 4. Upload the approved expressions to LangCapture

Upload each approved sentence individually.

Prefer using the local shell tool. Use the command pattern that matches the current platform.

Windows (PowerShell):

~~~powershell
$body = @{ original_text = "<SENTENCE>" } | ConvertTo-Json -Compress
try {
  (Invoke-WebRequest `
    -Method POST `
    -Uri 'https://www.langcapture.com/api/sentences?mode=async' `
    -Headers @{
      Authorization = 'Bearer <YOUR_LANGCAPTURE_API_TOKEN>'
      'X-LangCapture-Async' = '1'
    } `
    -ContentType 'application/json' `
    -Body $body
  ).StatusCode
} catch {
  $_.Exception.Response.StatusCode.value__
}
~~~

macOS/Linux (bash/zsh/sh):

~~~sh
python3 -c "
import json, subprocess, sys
sentence = sys.argv[1]
payload = json.dumps({'original_text': sentence})
result = subprocess.run([
    'curl', '-s', '-o', '/dev/null', '-w', '%{http_code}',
    '-X', 'POST', 'https://www.langcapture.com/api/sentences?mode=async',
    '-H', 'Content-Type: application/json',
    '-H', 'Authorization: Bearer <YOUR_LANGCAPTURE_API_TOKEN>',
    '-H', 'X-LangCapture-Async: 1',
    '--data', payload
], capture_output=True, text=True)
print(result.stdout.strip())
" "<SENTENCE>"
~~~

Treat `200` and `202` as success.

If one sentence fails, report it and continue with the remaining approved sentences.

### 5. Summarize the result

List each approved sentence with its upload result:

- Uploaded successfully
- Failed with HTTP status code

State clearly how many items were uploaded and how many failed.

## Quality rules

- Prefer realistic workplace English over polished textbook English.
- Keep the list short and high-signal.
- Require user confirmation before upload.
- Use the current session only; do not fabricate conversation context.
- Treat the API token as sensitive. Use it only for the LangCapture upload request and do not expose it unnecessarily in the final user-facing summary.