jiucai-capture

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.90). Yes — the skill actively fetches and scrapes user-generated content from the public site 九阳公社 (jiuyangongshe.com) via the study_hot / industry_chain / action scrapers and Playwright (see SKILL.md and a_stock_watcher/sources/* and content_fetcher.py), then the agent directly reads and AI-parses that untrusted content (ai_parser.parse_article) which influences filtering, database writes, and subsequent tool outputs.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). Flagging https://www.jiuyangongshe.com (and its runtime API endpoints such as /v1/timeline/news, /v1/industry/list, /action/field) because the skill's scrapers fetch JSON/article content from these URLs during runtime and directly inject that external content into the Gemini parsing prompt, allowing remote data to control the model's inputs.