when-agents-fail-comprehensive

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

• Potentially malicious external URL detected (high risk: 0.90). The skill contains a runtime fetch that pulls external prompt content via hub.pull("hwchase17/react"), which would load remote instructions that directly control the agent's prompts at runtime (flagged: hub.pull("hwchase17/react")); the arXiv link is only a static reference and not used at runtime.