The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The course provides templates for agents that process untrusted external data, such as customer emails and invoices, which represents an indirect prompt injection attack surface.
Ingestion points: SKILL.md (Chapter 4) describes workflows where agents read invoice lists and customer email content.
Boundary markers: The skill teaches the 'Role + Task + Constraints + Format' prompt pattern, which serves as a delimiter-based boundary to reduce accidental instruction following.
Capability inventory: The skill itself has no execution capabilities. It discusses platform tools like 'Code Interpreter' or 'Analysis Mode' which operate in sandboxed environments.
Sanitization: The course includes a dedicated chapter (Chapter 5) on security and privacy, providing explicit warnings about PII redaction and compliance with Israeli Privacy Protection Law Amendment 13.
[NO_CODE]: The skill consists entirely of Markdown educational material and JSON configuration/quiz data. There are no scripts (Python, JavaScript, Shell) or binaries included in the package.
[SAFE]: All external resource references, such as the israeli-personal-assistant skill and israeli-bank-scrapers mention, are legitimate tools within the vendor's ecosystem or well-known open-source projects. The documentation follows security best practices by warning users against hardcoding secrets and providing a 'never-paste' list for sensitive data.

ai-agents-for-everyone