outreach-resolve

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

• Third-party content exposure detected (medium risk: 0.30). In outreach-resolve (operator stage runner), the agent reads the sourced “segment” from Twenty (run’s ingested but not yet resolved leads) and then resolves/merges per contact via twenty-engine-sync, so outsider-authored lead data entering Twenty can be ingested and used at runtime.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill documentation invokes "npx -y @bbengamin/paperclip-mcp-server", which runs remote npm package code at runtime as the MCP fallback and thus executes external code the skill can rely on.