The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The tutorial provides multiple examples of RAG systems that ingest external text files and search results into LLM prompts.
Ingestion points: External knowledge base files (knowledge_base.txt), web search results in the SearchAgent, and product inventory data in the shopkeeper project.
Boundary markers: The RAG implementation uses simple delimiters like 上下文：\n{context}, but lacks explicit instructions to the model to ignore embedded commands within the retrieved text.
Capability inventory: The agents have capabilities for network requests via requests, SQL generation for database queries, and multi-step reasoning through LangGraph.
Sanitization: The snippets show basic prompt engineering for instruction following but do not demonstrate input sanitization or validation of the retrieved content before interpolation.
[COMMAND_EXECUTION]: The skill demonstrates NL2SQL (Natural Language to SQL) generation in the 'Shopkeeper Agent' project. Generating database queries directly from user input without demonstrating parameterization or strict validation is a common security pitfall that can lead to SQL injection if applied to production environments.
[EXTERNAL_DOWNLOADS]: The setup instructions include cloning external repositories (github.com/didilili/ai-agents-from-zero, github.com/langgenius/dify) and installing various libraries from the Python Package Index. While these are standard tools for the tutorial's domain, users should verify the integrity of external scripts before execution.

ai-agents-from-zero-tutorial