Analyzing AgentScope Library

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

• Third-party content exposure detected (high risk: 0.75). The required workflow instructs cloning and then using cat/ls to read the agentscope GitHub repo’s examples and docs/tutorials at runtime, which are outsider-authored free-text sources (public web content fetched at runtime) that can be 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 explicitly instructs at runtime to git clone https://github.com/agentscope-ai/agentscope to fetch repository files that will be read and used to guide answers, meaning external content is fetched during runtime and can directly control the agent's prompts/behavior.