kaggle-learner

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

• Third-party content exposure detected (high risk: 0.85). The required workflow involves ingesting competition-winning solution writeups (“前排方案详细技术分析”/code/templates) that are extracted from Kaggle competition pages at runtime and then placed into the skill’s knowledge markdown files; this is outsider-authored free text coming from Kaggle (public web content) into the LLM context via the kaggle-miner extraction + knowledge base update path.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill includes runtime web-scraping code that fetches AoPS content from https://artofproblemsolving.com and then injects that fetched content into LLM prompts (build_openmath_dataset -> llm_extract/generate_tir_from_solution), so the external URL is used at runtime to control prompts.