clay-core-workflow-b

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from external websites via Claygent research, which is then interpolated into prompts for generating email copy.
  • Ingestion points: In SKILL.md (Step 3), the prompt incorporates research data such as {{Recent News}} and {{Tech Stack}} derived from external websites.
  • Boundary markers: The instructions in SKILL.md (Step 3) use clear delimiters like 'Context about the prospect:' and 'Rules:' to help differentiate between data and instructions.
  • Capability inventory: The workflow focuses on text generation and data export via HTTP requests to a well-known service (Instantly.ai), without local code execution or sensitive file system access.
  • Sanitization: A QA script provided in SKILL.md (Step 4) checks for unresolved template variables and prohibited phrases, which helps mitigate accidental instruction obedience.
  • [SAFE]: The skill exports campaign-ready data to an official API endpoint for a well-known outreach service.
  • Evidence: Step 5 in SKILL.md provides configuration for sending data to https://api.instantly.ai/api/v1/lead/add using a placeholder for the authorization token ({{Instantly API Key}}).
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 03:46 AM