The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the web, which could be used to manipulate agent behavior via indirect prompt injection.
Ingestion points: distribb_research.py fetches text content from arbitrary websites found during its research phase; news_topics.py retrieves headlines and metadata from public Google News RSS feeds.
Boundary markers: The system prompts in distribb_research.py and distribb_writer.py use natural language instructions to limit the AI's output to the provided source data, but they lack robust delimiters or cryptographic signatures to separate external content from system-level instructions.
Capability inventory: The skill possesses extensive capabilities that could be abused if an injection were successful, including the ability to write local files, perform network requests to non-whitelisted domains, and publish content to connected external CMS platforms (WordPress, Shopify, etc.) through the Distribb API.
Sanitization: The skill uses the BeautifulSoup library to remove HTML tags from scraped data, but it does not perform advanced sanitization, validation, or escaping of the resulting text content before it is used in prompt interpolation.

distribb