agent-university

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

• Third-party content exposure detected (high risk: 0.75). The required workflow calls hosted MCP tools (e.g., agent_university_how_to_build, agent_university_get_artifact, agent_university_get_code_sample) that return “verbatim source content” from the Agent University corpus, which is outsider-authored public build documentation fetched at runtime and then ingested into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill registers and calls a hosted MCP endpoint (https://agent-university-api-production.up.railway.app/v1/mcp) at runtime to fetch full build guides and tooling instructions that are injected into the agent workflow, so the external content can directly control prompts/instructions and is a required dependency.