langchain

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The content intentionally exposes remote code execution vectors (an eval-based tool and a Python REPL tool) that can be invoked by agents/LLMs, creating a high-risk backdoor/abuse path for code execution and potential data exfiltration, though there are no obfuscated payloads or explicit hidden exfiltration endpoints.

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

• Third-party content exposure detected (high risk: 0.90). The skill registers and uses web-facing components—specifically DuckDuckGoSearchRun as a search tool in the agent (and WebBaseLoader referenced in the RAG loaders)—so the agent fetches and consumes open web content as part of its workflow, allowing untrusted third-party pages to influence decisions and actions.