bilibili-toolkit

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

• Third-party content exposure detected (high risk: 0.90). The skill actively fetches and ingests user-generated content from public Bilibili pages/APIs (e.g., scripts/bili_collect_and_export.py using bilibili_api and downloading audio, and scripts/bili_video.py scraping https://www.bilibili.com), then indexes those transcripts (scripts/bili_kb_llama.py) and uses them to drive LLM answers and reranking (scripts/bili_search_llama.py and bili_up_summarizer.py), so untrusted third‑party content is read and can materially influence model outputs and actions.