The Agent Skills Directory

[SAFE]: The skill provides structured technical instructions for designing complex architectures using established mathematical models such as Little's Law and GPU VRAM capacity formulas. The provided Go code for shuffle sharding is a standard implementation using safe, built-in libraries.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process technical inputs (NFRs, traffic projections, and model specs) to produce architecture designs. While it lacks explicit boundary markers for these inputs, it includes proactive security guardrails (ASI05 RCE Guard) that explicitly instruct the agent to avoid constructing infrastructure-as-code from unvalidated external content, effectively mitigating common injection risks.
[COMMAND_EXECUTION]: The skill references standard tools for file management (read_file, write_file) and development workflows (run_build, execute_command) within the platform's allowed configuration, solely for the purpose of generating and validating infrastructure specifications.

system-design