The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process user-supplied identity information, such as legal names and Taxpayer Identification Numbers (TINs). This data is used to generate tax drafts and guide the agent in interacting with external browser-based services like DocuSign or vendor portals. This ingestion creates a surface for indirect prompt injection where malicious input could attempt to influence the agent's behavior during sensitive operations.
Ingestion points: User-provided inputs for legal name, business name, TIN, and address as defined in SKILL.md and its associated workflow steps.
Boundary markers: The skill employs a structured workflow and a dedicated 'Validation' section in SKILL.md to ensure data matches expected patterns, though it does not specify character-level sanitization.
Capability inventory: The agent is instructed to use browser automation tools (e.g., Playwright) to fill forms on third-party sites and handle sensitive SSN/EIN data.
Sanitization: Validation logic exists to verify the format of TINs (9 digits, correct prefixes), which serves as a basic level of input filtering.
[DATA_EXFILTRATION]: The primary purpose of the skill involves the handling and transmission of sensitive Taxpayer Identification Numbers (SSNs/EINs). While the instructions in filing.md emphasize secure delivery through well-known services (DocuSign, Adobe Sign) and explicitly prohibit plain-text email or logging of sensitive data, the processing and transmission of this PII to external platforms constitutes a data exposure risk that must be managed by the host environment.

form-w9