The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes untrusted article text without boundary markers or sanitization. A malicious article could contain hidden instructions that manipulate the claim extraction process or force the agent to perform specific searches through Gemini's grounding tool.
Ingestion points: The run_fact_check function in scripts/fact_checker.py ingests external text from files using Path(input_path).read_text().
Boundary markers: None. The {article} content and {search_context} (which contains external search results) are interpolated directly into prompts without protective delimiters or instructions to ignore embedded commands.
Capability inventory: The skill can read local files, execute web searches via the Gemini google_search tool, and write Markdown files to the disk.
Sanitization: There is no escaping or validation logic applied to the input text before it is sent to the LLM.

fact-checker