deep-study

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

• Third-party content exposure detected (high risk: 1.00). The skill's Workflow (step 3 under "Walk" in SKILL.md) explicitly instructs the agent to use WebFetch (or resolve Context7 IDs) to fetch cited URLs from topic reports in .plans/research-with-instructions/*/REPORT.md, meaning it will read and act on arbitrary third‑party web content that can influence its findings.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly performs WebFetch on URLs cited in topic reports (the URLs listed in .plans/research-with-instructions/*/REPORT.md) at runtime and injects that fetched content into the agent's context to drive paraphrases, so arbitrary external URLs can directly control the agent's outputs.