list-cleaning
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, including lead lists (CSV/URL), LinkedIn profiles, and website content fetched via a browser. This creates an attack surface for indirect prompt injection where malicious instructions could be embedded in the data to influence the agent.
- Ingestion points:
SKILL.md(lead lists, URLs), LinkedIn profiles, and company websites. - Boundary markers: The skill explicitly references
references/agent-security.mdwhich mandates treating all fetched content as data, never as instructions, and requires quoting injection attempts rather than acting on them. - Capability inventory:
SKILL.mdutilizes thebrowsertool (Playwright) for live verification and interacts with external APIs (e.g., Prospeo) via MCP. - Sanitization:
references/agent-security.mdprovides strict rules for HTML escaping interpolated values and redacting tokens from output. - [COMMAND_EXECUTION]: The skill uses a headless browser (Playwright) to perform live verification of roles and companies. While this is an intended function for the skill's purpose, it represents a significant capability that is mitigated by the instructions in
references/agent-security.mdto avoid following instructions found within retrieved content. - [EXTERNAL_DOWNLOADS]: The skill fetches external data from user-provided URLs and integrates with third-party verification services (e.g., Prospeo) to validate email deliverability. These operations are scoped to the primary purpose of the skill and governed by internal compliance and security rules.
Audit Metadata