Fix interpreter escaping issue

[Ynewtime]

Hi,

issue

when using interpreter to optimize markdown content, if \ backslash or characters to be escaped are present, the output from LLM is not correctly escaped. e.g.:

Input content from web page:

# bash
export no_proxy="$NO_PROXY"

# powershell
.\Export-WindowsCerts.ps1

Write-Host "`n Export successfully" -ForegroundColor Green

Output content from LLM:

```bash
export no_proxy=\\"$NO_PROXY\\"
```

```powershell
# powershell
.\\Export-WindowsCerts.ps1

Write-Host \\"`n Export successfully\\" -ForegroundColor Green
```

You can reproduce this with this page: https://telegra.ph/obsidian-clipper-interpret-issue-reproduction-12-15

with this template:

interpret-test-clipper.json

or simply use this prompt for the content:

{{"better formatting markdown content in GFM styles. FORMAT ONLY, DO NOT MODIFY CONTENT!!!"}}

Basically this issue is provider- and model- agnostic, I tested this with most of the popular models including:

• google/gemini-2.5-flash-lite
• deepseek/deepseek-v3.2
• anthropic/claude-haiku-4.5
• x-ai/grok-code-fast-1

(as per my test, google/gemini-2.5-flash-lite is the best model in this scenario, fast and good enough)

solution

I tried to fix this issue with Claude Opus 4.5, and it was fixed in just one shot—worked really well. I reviewed the code and verified it by building, installing it in my browser, and testing it on hundreds of web pages.

Hence this PR.

You can also try the built extention here:

obsidian-web-clipper-0.12.0-chrome.zip