BUG: Interpreter returns raw prompt text instead of processed output — Claude Sonnet and Haiku only (Opus works)

[janpajerski]

Version

• OS: Windows 10
• Browser Chrome
• Web Clipper version: 1.2.1
• Obsidian version: 1.12.7

Describe the bug

When using the Interpreter with any Claude model except Opus 4.5, the note is created in Obsidian but contains the raw prompt text instead of the interpreted result. Opus 4.5 works correctly — it takes 30-40 seconds and produces a properly filled note. All other models (Sonnet 4.5 preset, Haiku 4.5 preset, and Sonnet 4.6 custom) also take 30-40 seconds but create a note containing only the raw prompt.
No errors appear in the extension's background service worker console. No network request to api.anthropic.com is visible in the DevTools Network tab for any model (including Opus), suggesting API calls are made from the service worker context.

The Anthropic API key is valid — Tier 1 account with available credits, and Sonnet/Haiku are listed as active in the console rate limits page.

This may be a regression of issue #321.

Expected behavior

Sonnet and Haiku should return interpreted output, same as Opus.

URLs where the bug occurs

Is affecting all URLs tested.

To reproduce

1. Open webpage
2. Click on the Obsidian Web Clipper and select a template that uses an interpreter.
3. Select Claude Sonnet 4.5 (preset) as the Interpreter model
4. Click on "Add to Obsidian" button.
5. Note is created with raw prompt text instead of interpreted output
   6 Switch to Claude Opus 4.5 (preset) follow the same steps and interpreted output is added to Obsidian as expected.

Your template file

If you are using a custom template (i.e. not the Default template). Go to Web Clipper settings and click More → Copy as JSON. Paste the JSON code below.

{
  "schemaVersion": "0.1.0",
  "name": "Default - Summarized content",
  "behavior": "create",
  "noteContentFormat": "## My Summary\n\n{{\"Summarize the following webpage into a structured Obsidian note in Markdown format. Use H2 and H3 headers to organize the content logically. Ignore filler content such as sponsor segments, social media calls to action, and generic introductions.\n\nOutput ONLY the following sections in this exact order. Use Obsidian callout syntax where appropriate: \n> [!important]\n> For critical concepts or must-remember ideas.\n> [!note]\n> For notable quotes or key statements from the speaker.\n> [!question]\n> For thought-provoking questions raised by the content.\n> [!example]\n> For concrete examples or case studies mentioned.\n\n## Overview\nProvide a concise summary of the webpage in natural language.\n\n## Key Ideas\n3–7 distilled insights from the webpage. You can add more, or use less, when appropriate.\n\n## Practical Takeaways\n3–5 actionable steps the reader can apply immediately. Add more, or use less, when appropriate.\n\n## Anki Flashcards\nFor each key concept, generate a flashcard in the exact format below. \n\nTARGET DECK: ***Enter Deck***\nSTART\nBasic-Obsidian\nFront: [Question about the concept]\nBack: [Concise answer]\nTags: ***Enter Tags***\nEND\n\n## Related Concepts\nA list of topics, books, or frameworks mentioned or implied in the content of the webpage.\n\nWebpage content:{content}\"}}",
  "properties": [
    {
      "name": "tags",
      "value": "",
      "type": "multitext"
    },
    {
      "name": "title",
      "value": "{{title}}",
      "type": "text"
    },
    {
      "name": "source",
      "value": "{{url}}",
      "type": "text"
    },
    {
      "name": "author",
      "value": "{{author|split:\\\", \\\"|wikilink|join}}",
      "type": "multitext"
    },
    {
      "name": "published",
      "value": "{{published}}",
      "type": "date"
    },
    {
      "name": "created",
      "value": "{{date}}",
      "type": "date"
    },
    {
      "name": "description",
      "value": "{{description}}",
      "type": "text"
    }
  ],
  "triggers": [],
  "noteNameFormat": "{{selector:h1}}",
  "path": "00. Inbox"
}