outreach-source

Pass

Audited by Gen Agent Trust Hub on Aug 4, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill requires the installation of the paperclipai CLI and utilizes the @bbengamin/paperclip-mcp-server via npx. Both are vendor-specific resources from the skill's author for interacting with the Paperclip ecosystem.\n- [COMMAND_EXECUTION]: The skill uses the paperclipai CLI and Model Context Protocol (MCP) tools to perform administrative tasks, such as managing authentication, context settings, and interacting with Paperclip API resources.\n- [DATA_EXFILTRATION]: The skill is designed to transfer lead data from external sources (Apollo, Grinfi) into the Twenty CRM. It interacts with local authentication stores (~/.paperclip/auth.json) to manage its connection to the Paperclip platform, which is consistent with the intended operation of the CLI tool.\n- [PROMPT_INJECTION]: The skill processes lead data from external providers (Apollo search results) and imported files (CSVs), creating an attack surface for indirect prompt injection.\n
  • Ingestion points: Data enters the system via Apollo search tools (APOLLO_PEOPLE_SEARCH) and file imports (grinfi__import_leads_from_file).\n
  • Boundary markers: The skill relies on structured ingestion via the twenty-engine-sync tool but lacks specific prompt-level delimiters to neutralize potential instructions embedded within the sourced lead data.\n
  • Capability inventory: The agent can write to CRM systems, execute shell commands through the Paperclip CLI, and perform network requests to various service providers.\n
  • Sanitization: Explicit sanitization logic for the lead content is not detailed in the provided instructional files.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 4, 2026, 08:27 PM
Security Audit — agent-trust-hub — outreach-source