agent-reach

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

• Third-party content exposure detected (high risk: 0.85). At runtime the skill installs and then uses Agent Reach to “read and search” user-specified URLs/queries via backends like Jina Reader/Exa/MCP, which fetch and extract arbitrary public web content (outsider-authored text) 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 instructs installing and running remote code at runtime via pipx/pip from https://github.com/Panniantong/agent-reach/archive/main.zip (and related GitHub URLs), which downloads and executes upstream code that the skill requires—this is a runtime external dependency that executes remote code.