The Agent Skills Directory

[SAFE]: The skill is focused on tax compliance and regulatory reporting. It provides a structured workflow for agents to assist users in filing Form 8300 with the IRS and FinCEN.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface as it processes user-provided transaction details and personally identifiable information (PII) to fill out tax forms. However, this is the primary intended function of the skill. The instructions include robust validation steps, field-by-field mapping, and specific warnings to the agent not to store sensitive data like SSNs or government ID numbers in long-term memory (logs or vector stores). This significantly mitigates the risk of instructions embedded in data influencing the agent's behavior.
[DATA_EXFILTRATION]: No unauthorized data exfiltration patterns were detected. All network references and filing instructions target official government domains (irs.gov, fincen.treas.gov) or the vendor's own official resources (jupid.com). The skill explicitly warns against transmitting sensitive data via unencrypted channels like email or SMS.
[COMMAND_EXECUTION]: The skill describes a deterministic browser automation workflow (e.g., via Playwright or Puppeteer) for form submission on the official FinCEN portal. This is a legitimate extension of agent capabilities for the specified task and does not involve arbitrary or malicious shell command execution.

form-8300