The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's data retrieval scripts fetch JSON content from https://newsnow.busiyi.world/api/s. This external domain is not associated with a trusted vendor or well-known service provider, posing a supply-chain risk for the data being processed.
[PROMPT_INJECTION]: There is a documented surface for indirect prompt injection (Category 8) because the skill ingests headlines from external news platforms and presents them to the LLM for analysis without protective delimiters.
Ingestion points: News titles are retrieved by scripts/fetch_newsnow_topics.py and scripts/fetch_hot_topics.py from external platform APIs.
Boundary markers: The prompts defined in SKILL.md do not utilize boundary markers or explicit instructions to treat the ingested news titles as data rather than instructions.
Capability inventory: The skill includes scripts capable of network communication (requests) and local file persistence (json.dump), which could be leveraged if an injection is successful.
Sanitization: The skill does not implement any filtering or sanitization of the news titles before they are included in the prompt context.
[DATA_EXFILTRATION]: The skill initiates network requests to external domains (newsnow.busiyi.world, weibo.com, zhihu.com) that are not on the designated safe list, which is characteristic of potential data exposure vectors.

capital-market-topic-scout