paper2code

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

• Third-party content exposure detected (high risk: 1.00). The skill explicitly fetches and ingests open web content (e.g., SKILL.md "For arXiv URL: curl -L 'https://arxiv.org/pdf/....pdf' -o paper.pdf" and 05_reference_search.md which directs web/GitHub searches and selecting repositories), and the agent is required to read and act on that untrusted, user-generated third‑party content as part of its implementation workflow, so external page/content could materially influence tool actions.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly downloads arXiv papers at runtime using curl (e.g., https://arxiv.org/pdf/xxxx.xxxxx.pdf), which are then parsed and injected into the agent’s processing pipeline to drive implementation instructions, so external content can directly control the agent's prompts/behavior.