Interpreter fails silently on multi-line prompt variables

[maciejklimek]

Description

Prompt variables containing newlines and bullet points fail silently — the raw prompt text appears in the output instead of the LLM response. Shorter single-line prompts in the same template work correctly.

Reproduction

1. Create a template with a multi-line prompt variable in noteContentFormat (containing line breaks and bullet points)
2. In the same template, add a second shorter single-line prompt
3. Clip any page and run the interpreter
4. The first (multi-line) prompt appears as raw text in the resulting note
5. The second (single-line) prompt is processed correctly and shows the LLM response

Expected behavior

Both prompts should be sent to the LLM and replaced with responses.

Workaround

Keep prompts on a single line without newline characters inside the prompt variable.

Environment

• Web Clipper version: 0.12.x
• Browser: Safari
• Model: Claude 3.5 Haiku (Anthropic)

[maciejklimek]

Here's the exact template that reproduces the issue:

{
  "schemaVersion": "0.1.0",
  "name": "GitHub Repo",
  "behavior": "create",
  "noteContentFormat": "## Relevance to My Work\n\n{{\"Analyze this GitHub repository and explain how it could be useful for someone who:\n- Works on Speech AI at Google (Cloud AI Speech team)\n- Develops voice-driven programming tools (like SpeakEasy - a voice coding platform)\n- Works primarily with Python\n- Builds voice interfaces and speech recognition systems\n- Is interested in AI-assisted development workflows and MCP integrations\n\nBe specific about:\n1. Which aspects of my work this could enhance\n2. Concrete use cases or integration ideas\n3. Any limitations or considerations\n\nIf the repo is not relevant to these areas, say so briefly and note what it IS useful for instead.\"}}\n\n---\n\n## Repository Overview\n\n{{\"Provide a concise summary: what this repo does, key features, tech stack, and maturity/maintenance status.\"}}\n\n---\n\n## Raw Content\n\n{{content}}",
  "properties": [
    {"name": "title", "value": "{{title}}", "type": "text"},
    {"name": "source", "value": "{{url}}", "type": "text"},
    {"name": "stars", "value": "{{selector:#repo-stars-counter-star}}", "type": "text"},
    {"name": "created", "value": "{{date}}", "type": "date"},
    {"name": "tags", "value": "clippings, github", "type": "multitext"}
  ],
  "triggers": ["https://github.com"],
  "noteNameFormat": "GH - {{title}}",
  "path": "Clippings/GitHub",
  "context": "{{content}}"
}

The first prompt variable (the long multi-line one with bullet points) appears as raw text in the output, while the second prompt ("Provide a concise summary...") works correctly and returns the LLM response.