The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements logic for web scraping, which involves ingesting untrusted data from the internet. This presents a surface for indirect prompt injection attacks.
Ingestion points: Data is ingested through browser navigation and extraction logic in SKILL.md and references/advanced.md.
Boundary markers: The extraction logic lacks explicit boundary markers or instructions to the agent to treat scraped content as data only, increasing the risk that the agent might follow instructions found within the scraped HTML.
Capability inventory: The skill demonstrates the ability to perform network requests and write results to the local file system using the saveResults function in references/advanced.md.
Sanitization: While the schema includes a transform function for data normalization, it does not include sanitization layers designed to detect or strip malicious prompt-injection patterns from the extracted text.

scraping-automation