aris-autonomous-ml-research

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

• Third-party content exposure detected (high risk: 0.90). The SKILL.md explicitly instructs the agent to fetch and parse open web content (e.g., "Discovers related papers from arXiv" and cloning/reading GitHub repos in the /research-pipeline and example workflows), and then reads/interprets that untrusted, user-provided content to drive experiments, code changes, reviews, and rebuttals—allowing third-party material to influence tool use and decisions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly performs runtime git clones and runs code from external repositories (e.g., https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep.git and example experiment repo https://github.com/kimiyoung/transformer-xl), so fetched remote content is executed/used at runtime and therefore directly controls agent behavior.