The Agent Skills Directory

[SAFE]: The skill's primary focus is on engineering and architecture for LLM systems. All provided scripts and documentation are consistent with this purpose and follow standard industry patterns.\n- [PROMPT_INJECTION]: Static analysis flagged the string 'ignore previous instructions' in REFERENCE.md. However, context reveals this is part of an 'injection_patterns' list within a 'check_safety' function. This is a security feature intended to detect and block malicious input, not an attempt to subvert the agent.\n- [CREDENTIALS_UNSAFE]: No hardcoded secrets or API keys were found. Code samples use standard placeholders like 'YOUR_API_KEY' or empty strings, which is considered safe practice for documentation.\n- [COMMAND_EXECUTION]: The skill provides Python scripts for benchmarking, evaluation, and serving models. These involve standard library usage (FastAPI, Transformers, etc.) and do not contain arbitrary or hidden command execution logic.\n- [EXTERNAL_DOWNLOADS]: The skill references standard machine learning and LLM libraries (Hugging Face, OpenAI, Pinecone, etc.). These are well-known, trusted services in the AI ecosystem and are necessary for the skill's functionality.

llm-architect