outreach-enrich

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

• Third-party content exposure detected (medium risk: 0.30). The enrich stage runs per-lead by reading the “next unhandled lead” from Twenty (per-lead progress stored in Twenty, including enrichment status), meaning outsider-submitted lead data can be ingested into the runtime without selecting a specific item beyond choosing the next lead.

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

• Potentially malicious external URL detected (high risk: 0.70). The docs/configs include the command "npx -y @bbengamin/paperclip-mcp-server" which will fetch and execute an npm package at runtime to provide an MCP server used by skills (i.e., it runs remote code during skill operation).