The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The classify_leads_llm.py script interpolates untrusted data scraped from external sources into an LLM prompt. This creates a surface where adversarial text in a company description could trick the LLM into incorrect classification. Ingestion points: Lead data is read from .tmp/leads.json in the main function of classify_leads_llm.py. Boundary markers: No delimiters or isolation instructions are present in the classification prompt to distinguish between system instructions and lead data. Capability inventory: The classification output only affects a label in the output data and Google Sheet; it does not trigger code execution or network operations. Sanitization: Scraped data is used directly in the prompt without sanitization or escaping.

scrape-leads