brex-automation

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

• Third-party content exposure detected (high risk: 0.80). The skill requires calling RUBE_SEARCH_TOOLS against an external MCP server (e.g., https://rube.app/mcp) to fetch tool schemas, recommended execution plans, and runtime data that the agent must read and act on, exposing it to untrusted third-party content that could carry indirect prompt injections.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill instructs adding the MCP server endpoint https://rube.app/mcp which is contacted at runtime (via RUBE_SEARCH_TOOLS / RUBE_MULTI_EXECUTE_TOOL) to fetch tool schemas and execute tools, so remote content from that URL directly controls prompts/instructions and remote execution.

MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).

• Direct money access detected (high risk: 1.00). The skill is explicitly for "Automate Brex operations" and requires connecting to the Brex toolkit via RUBE_MANAGE_CONNECTIONS (toolkit "brex") and then executing discovered tool slugs with RUBE_MULTI_EXECUTE_TOOL. Brex is a bank/financial platform, and the described workflow is clearly designed to discover and invoke Brex-specific API/tool schemas (which would include transactional/financial operations such as transfers, payments, card/account management). This is not a generic browser or HTTP caller: it instructs use of a specific financial toolkit and executing its tools, which constitutes direct financial execution capability.