coda-automation

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow reads Coda document/page content and table rows via tools like CODA_LIST_PAGES, CODA_GET_A_PAGE, CODA_LIST_TABLE_ROWS, CODA_SEARCH_ROW, and CODA_GET_A_ROW, where any outsider who can create or share content into a document the user connects to can supply the free text that the agent ingests.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill requires adding the MCP endpoint https://rube.app/mcp and instructs the agent to call RUBE_SEARCH_TOOLS to fetch current tool schemas at runtime, meaning remote content from that URL will directly control available prompts/instructions and tool behavior.