cold-email-outreach
Pass
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests external data from company websites, LinkedIn profiles, and news articles to facilitate target research and email personalization. This creates a surface for indirect prompt injection where untrusted external content could attempt to influence the agent's instructions. Ingestion points include target company research and contact profiles as defined in SKILL.md and references/templates/research-report.md. Boundary markers are strong, with mandatory human-in-the-loop review required for all research reports and email drafts before any action is taken. Capability inventory shows the skill is limited to generating text-based artifacts and managing local configuration files; it lacks autonomous execution for sending emails or accessing sensitive credentials. Sanitization relies on the agent's explicit instructions to flag missing evidence and the user's final factual approval.
- [DATA_EXPOSURE]: The skill initializes and writes to a state manifest and campaign configuration files in the local project directory (e.g., .cold-email-outreach/state-manifest.json). While the skill accesses the filesystem, references/bootstrap/bootstrap.md and references/bootstrap/user-welcome.md contain clear safety directives forbidding the storage of secrets, API keys, or email credentials in these state files.
Audit Metadata