cold-email-starter-kit
Pass
Audited by Gen Agent Trust Hub on May 7, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill facilitates interaction with several third-party service providers via their official APIs. These include domain registration (Dynadot), email hosting (Zapmail), lead sourcing (Prospeo), and campaign management (Smartlead and Instantly). It also uses OpenRouter for AI-driven enrichment and RapidAPI for news and social data.
- [COMMAND_EXECUTION]: The toolkit provides a series of automation scripts for the Node.js environment. These scripts manage the full lifecycle of a cold email campaign, from domain generation and bulk purchasing to lead list exportation and campaign activation.
- [DATA_EXFILTRATION]: As part of its intended functionality, the skill reads API keys from a local
.envfile and transmits them to the authorized endpoints of the service providers. It also transfers lead data from CSV files to external platforms for enrichment and sending. - [PROMPT_INJECTION]: The
ai-company-analysis.tsscript presents an indirect prompt injection surface by interpolating data from lead CSV files directly into LLM prompts sent to OpenRouter. This could potentially allow for instruction-based manipulation if a lead entry contains malicious text. - Ingestion points: Lead data is read from CSV files like
leads.csvand used by multiple scripts. - Boundary markers: No explicit delimiters are used to separate lead-derived variables from instructions in the LLM prompt templates.
- Capability inventory: The skill uses network operations for API interaction, file-system access for credential and data management, and script execution for process automation.
- Sanitization: Lead data is not sanitized or escaped before being used in prompt interpolation.
Audit Metadata