outreach-execution
Pass
Audited by Gen Agent Trust Hub on Aug 10, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No malicious patterns or unauthorized execution methods were detected. The skill is primarily a template for instructional behavior, focusing on sender voice management and outreach orchestration with robust safety rails.
- [DATA_EXFILTRATION]: The skill is designed to ingest sensitive organizational data, including CRM contacts, deal history, and meeting transcripts (Step 5), to personalize outreach content. This access is inherent to the skill's purpose. Security risk is mitigated by the design requirement that the AI only 'stages' or 'queues' drafts for a human
{{APPROVAL_OWNER}}, preventing automated data exfiltration. - [PROMPT_INJECTION]: The skill ingests untrusted external data, such as LinkedIn profile activity and intent signals, which represents an indirect prompt injection surface. The skill instructions effectively mitigate this risk by mandating human review and prohibiting the auto-sending of AI-generated content to senior stakeholders at large organizations.
- [COMMAND_EXECUTION]: The skill uses placeholders for platform-specific tools (e.g.,
{{CRM}},{{RESOURCE_LIBRARY}}) but does not contain any direct shell command execution, subprocess spawning, or unauthorized privilege escalation attempts.
Audit Metadata