build-product-integrations

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

• Third-party content exposure detected (high risk: 0.90). The SKILL.md explicitly describes using Membrane actions as AI agent tools (see "AI Agent Tools" and "Dynamic Tools (runtime discovery)" which call GET /actions?connectionId=... and POST /actions//run), meaning the agent will fetch and interpret outputs from external apps (Slack, GitHub, Notion, etc.) that are user-generated/untrusted and can materially change agent behavior.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill uses the Membrane agent API at https://api.getmembrane.com (via agent-session endpoints and the MCP SSE URL https://mcp.integration.app/sse?...) at runtime to create agent sessions and receive generated connector/action specifications, meaning remote content from these URLs directly controls the agent's instructions/behavior and is relied on as part of the workflow.