The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions define a strictly restricted role for the agent, providing explicit negative constraints that prevent it from releasing payments or suggesting the override of compliance holds.\n- [COMMAND_EXECUTION]: The skill executes a specific Python script using a fixed path. Analysis of the script shows it uses only standard library modules and performs data processing without invoking shells or executing arbitrary commands.\n- [DATA_EXFILTRATION]: The Python script contains no network libraries or exfiltration patterns. It is limited to reading the local payment and sanctions files provided within the skill package.\n- [REMOTE_CODE_EXECUTION]: No remote dependencies or dynamic code loading mechanisms are used. The skill is self-contained and does not fetch external assets at runtime.\n- [EXTERNAL_DOWNLOADS]: The skill does not perform any downloads. The URL mentioned in the sanctions data fixture points to an official U.S. government domain (ofac.treas.gov), which is a trusted source for financial compliance data.

payment-screening