Skip to content

fact_checker.md

Latest commit

 

History

History
32 lines (22 loc) · 1.9 KB

fact_checker.md

File metadata and controls

32 lines (22 loc) · 1.9 KB
Raw

🕵️ Fact Checker Skill

The Fact Checker skill provides high-fidelity, investigative AI analysis for documents, URLs, and video content. It is designed to identify sensitive information, analyze innovation potential, and provide visual evidence of its reasoning.

🚀 Key Features

  • Sensitive Data Discovery: Automatically identifies claims related to legal conflicts, contract terminations, and sensitive corporate data.
  • Innovation Synthesis: Extracts the "Innovation Impact" and "Cognitive Potential" of the document's core concepts.
  • Visual Reasoning (VizHighlight): Automatically generates screen recordings (MP4) with visual highlighting of internal passages used as evidence.
  • Evidence Mapping: Links each finding to its source (URL or file path) and assigns an investigative probability score.
  • Investigative Summary: Synthesizes multiple findings into a coherent "Investigative Narrative".

🛠️ Typical Workflow

  1. Ingestion: Provide a source (URL or text).
  2. Analysis: Gemini 3 performs deep cognitive reasoning.
  3. Visualization: Automatically triggers the Photograph skill to record the screen while highlighting findings.
  4. Reporting: Saves a structured JSON report and the associated video evidence.

💻 Example Usage

# Analyze a confidential report and record the reasoning process
python3 .gemini/skills/fact-checker/scripts/fact_check_record.py "https://example.com/sensitive-report"

🎥 Visual Evidence (VizHighlight)

The tool doesn't just claim truth; it shows it. Every fact-check produces a visual artifact where the agent literally highlights the evidence on the screen while explaining its reasoning.

🛡️ Attribution & DID

All findings are attributed to the agent's DID (did:oyd:zQmcVHWDMeXtj273A9gNAnEG2EdrGEjtQiFuw9PncyVgs9z), ensuring accountability and provenance in the Croissant ecosystem.