The Agent Skills Directory

[SAFE]: The skill provides architectural guidance for designing secure system prompts and tool schemas, emphasizing the use of explicit guardrails and confirmation policies to prevent unauthorized actions.
[SAFE]: The provided Python script scripts/prompt_optimizer.py is a static analysis tool that operates locally on text inputs to estimate token counts and identify vague instructions; it uses standard libraries and performs no network operations.
[SAFE]: The documentation in references/mcp-engineering.md includes security best practices, such as mandatory DNS rebinding protection for local MCP servers and the implementation of approval handlers for sensitive operations.
[INDIRECT_PROMPT_INJECTION]: The skill documents techniques for injecting memory and runtime context into prompts (e.g., references/system-prompt-architecture.md). It provides a framework for managing this surface safely through delimiters, instruction forwarding, and sanitization rules.

prompt-engineering