The Agent Skills Directory

[SAFE]: The skill implements robust security templates, including a PromptInjectionGuard that filters for common injection keywords and enforces a 4,000-character input limit for untrusted data.\n- [SAFE]: Generated code templates utilize XML-style boundary markers (e.g., <ticket>) to encapsulate untrusted user input, effectively isolating it from system instructions to prevent prompt injection.\n- [SAFE]: The scaffolding logic promotes secure error handling by ensuring that internal system details and stack traces are logged server-side and never returned to the end-user, mitigating the risk of information disclosure.\n- [SAFE]: The skill relies on authorized Model Context Protocol (MCP) tools for project creation and documentation retrieval, ensuring a controlled and verifiable workflow within the agentic environment.

scaffold-project