The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data (e.g., personal details for form filling) and interacts with external government portals via WebFetch and Playwright. This creates an attack surface for indirect prompt injection, where malicious content embedded in user data or on a government page could attempt to influence the agent's behavior.
Ingestion points: User-supplied form data and dynamic content from gov.il portals.
Boundary markers: The skill relies on natural language instructions to separate data from instructions, which is a standard but non-robust boundary.
Capability inventory: The skill has access to Bash(python:*), Bash(pip:*), and WebFetch tools, which could be leveraged if an injection is successful.
Sanitization: The skill provides validation functions for specific formats (ID numbers, phone numbers, IBANs) which provides some level of data integrity.
[COMMAND_EXECUTION]: The skill instructions utilize Bash(python:*) to execute a local helper script (scripts/fill_form.py) for data validation. This is a standard use of the available tools for the skill's primary purpose.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of several external Python packages including playwright, pikepdf, and pypdf via pip. It also involves downloading Chromium binaries through the Playwright CLI. These are well-known and legitimate tools from established sources.

israeli-gov-form-automator