The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes untrusted data from external sources, including spreadsheet uploads and third-party API payloads from Gusto and QuickBooks Payroll, as outlined in the timesheet intake reference. While the instructions do not specify explicit boundary markers or sanitization for this ingested content, the risk is mitigated by the skill's core design. The capability inventory—which includes sensitive operations such as updating payroll records and network fetches—is governed by a mandatory human-in-the-loop approval gate. Every anomaly detected in the data must be manually resolved by the owner, preventing automated execution of potentially malicious instructions embedded in input files.
Sensitive Data Privacy: The skill interacts with sensitive employee records and financial data. It contains explicit directives for the agent to avoid reproducing or disclosing highly sensitive personal identifiable information (PII) such as Social Security Numbers, dates of birth, or banking details. This approach follows best practices for data minimization and privacy.
External Connection Management: The skill provides guidance for extending its functionality to additional tools through a connector-building process. This process prioritizes established connection directories and well-known services like Zapier, avoiding unverified third-party API interactions.
Tool Access Constraints: The skill metadata defines a limited set of allowed tools, restricting the agent's execution environment. While the instructions describe a variety of payroll-specific functions, these operations are scoped within the provided configuration to ensure focused and authorized tool usage.

payroll-prep