The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes news content fetched from external websites (Wall Street CN and 36kr). This creates a surface for indirect prompt injection if malicious instructions are embedded in news headlines or article text.
Ingestion points: Data is fetched via scripts/fetch_china_macro_news.py from api-one.wallstcn.com and 36kr.com (documented in references/data-sources.md).
Boundary markers: The agent is instructed to summarize and categorize the JSON output, but there are no explicit instructions to ignore embedded commands within the fetched content field.
Capability inventory: The agent is instructed to display content to the user and save reports to the local reports/ directory.
Sanitization: The fetch_url_content function in scripts/fetch_china_macro_news.py uses BeautifulSoup to remove script, style, nav, footer, and header elements, and truncates text to 3000 characters, which significantly reduces the attack surface.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to fetch news data from external financial news providers.
Evidence: The script scripts/fetch_china_macro_news.py uses the requests library to access https://api-one.wallstcn.com/apiv1/content/information-flow and https://36kr.com/newsflashes.

list-china-today-macro-news