Fact-Check: Multimodal News Verification Skill

When to Use

• User asks to verify a news article, claim, tweet, or social media post
• User asks "is this true?" about any statement
• User mentions fake news, misinformation, or disinformation
• User provides a URL and asks to check its truthfulness
• User asks to verify an image (is it real, AI-generated, photoshopped, manipulated, out-of-context)
• User asks to verify a video (is it deepfake, manipulated, real footage)
• User shares an image or video and asks if it is authentic
• User asks to fact-check content in any language

Stage 0: Input Parsing

1. Detect input modality

Determine what the user provided. Check in this order:

• Image file: The user attached or provided an image (JPEG, PNG, WebP, GIF, etc.). Set modality = image.
• Video file: The user attached or provided a video (MP4, MOV, AVI, WebM, etc.). Set modality = video.
• URL: The input starts with http:// or https://.
  • If the URL points directly to an image file (ends in .jpg, .png, .webp, etc.) → download it and set modality = image.
  • If the URL points directly to a video file (ends in .mp4, .mov, .webm, etc.) → download it and set modality = video.
  • Otherwise → use WebFetch to retrieve the page content. Extract only the main article body. Set modality = text.
• Plain text: None of the above. Set modality = text.

If the user provides both an image/video AND a text claim (e.g., "Is this photo from the 2024 earthquake?"), record both. Both branches will run and results will be combined in the report.

2. Detect language

Identify the language the user is writing in. This determines the report language and search query language.

3. Handle long text input (over 500 words / 1000 Chinese characters)

Only applies when modality = text.

• Ask the user: "This article is quite long. Would you like me to fact-check the entire article, or is there a specific claim you'd like me to verify?"
• 中文提示："这篇文章较长。你希望我核查全文，还是有某个具体声明需要验证？"

4. Route to pipeline

• modality = text → load and follow {baseDir}/references/text_pipeline.md
• modality = image → load and follow {baseDir}/references/image_pipeline.md
• modality = video → load and follow {baseDir}/references/video_pipeline.md

5. Store internally

input_text, language, modality (text/image/video), source_url (if URL), file_path (if local file).

Execution Logging

At the start of each stage/step, output a status line. Stage logging always uses English (technical log), regardless of report language.

[Stage N] Stage Name — status/result summary

Report Generation (Stage 5)

All pipelines end here. Load the report template from the pipeline-specific reference file.

Language rule