The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied Markdown files to extract and verify claims. A maliciously crafted document could contain instructions intended to override the agent's logic or bias its fact-checking results.
Ingestion points: The scripts/fact_checker.py script reads the entire content of the input Markdown file for processing.
Boundary markers: The prompts in scripts/fact_checker.py (such as SPLIT_PROMPT and VERIFY_PROMPT) interpolate document content directly into the instructions without using distinct delimiters or explicit warnings for the AI to ignore instructions embedded within the text.
Capability inventory: The script has the ability to read local files, write JSON results to the filesystem, and perform network requests via the Gemini API with Google Search capabilities.
Sanitization: There is no specific sanitization or filtering of the input text performed before it is passed to the LLM for analysis.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to interact with Google's Gemini API and utilizes the integrated Google Search tool.
Evidence: The script scripts/fact_checker.py uses the google-genai library to call model generation endpoints and enables the google_search tool for verification.
Context: These network calls target well-known services from Google and are essential for the skill's stated purpose of AI-powered fact-checking.

fact-checker