universal-research-orchestrator

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

• Third-party content exposure detected (high risk: 1.00). The skill explicitly requires web searches of public sources (see Phase 1 Agent Prompt Template "MUST DO: Search real sources: writeups, docs, CVEs, papers, GitHub issues, CTF solutions" and "When you find a promising URL, FETCH the entire resource") and Phase 5 archives via curl -sL <url>, meaning it ingests untrusted, user-generated public content that will be read and used to drive decisions and spawning of agents.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs agents to fetch arbitrary resources at runtime (e.g., "When you find a promising URL, FETCH the entire resource" and the archive step "curl -sL → sources//.md"), which means external content would be injected into the agent's context and can directly control prompts/outputs.