The Agent Skills Directory

[PROMPT_INJECTION]: The skill fetches news headlines, summaries, and social media posts from external platforms, which constitutes an indirect prompt injection surface. An attacker could potentially influence the agent's behavior by injecting instructions into a news story or a forum post processed by this skill.
Ingestion points: Public APIs and HTML pages from news providers (Sina, CLS, Wallstreetcn, Jin10, EastMoney) and investor forums (EastMoney Guba) in scripts/fetch_intel.py.
Boundary markers: Absent; the fetched content is formatted into JSON or Markdown without explicit delimiters or instructions to the LLM to ignore embedded commands.
Capability inventory: The skill's scripts do not possess dangerous capabilities such as arbitrary command execution, file system modification, or non-whitelisted network POST operations.
Sanitization: The skill performs basic sanitization by stripping HTML tags and unescaping entities using the _strip_html function in scripts/fetch_intel.py.
[EXTERNAL_DOWNLOADS]: The skill fetches data from well-known financial information services and government domains.
Evidence: Network requests are directed to official portals like csrc.gov.cn, pbc.gov.cn, sse.com.cn, and established platforms such as sina.com.cn, cls.cn, and eastmoney.com.
[SAFE]: Sensitive information management is handled correctly. The skill uses an environment variable (JIN10_APP_ID) for its only required API credential rather than hardcoding secrets.

ashare-news-fetcher