The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides documentation and code templates for building Retrieval-Augmented Generation (RAG) and tool-using agents, which are susceptible to indirect prompt injection from untrusted external data.
Ingestion points: templates/rag-pipeline.py and references/rag-strategies.md demonstrate ingesting data from external URLs using WebBaseLoader (e.g., https://example.com/docs).
Boundary markers: Prompt templates demonstrated in templates/rag-pipeline.py and references/rag-strategies.md do not include explicit delimiters or instructions for the LLM to ignore embedded commands within the retrieved context.
Capability inventory: Templates such as templates/agent-with-tools.py enable agents to call external tools (e.g., search_web), which could be triggered by adversarial instructions in retrieved content.
Sanitization: The provided examples do not demonstrate sanitization or validation of documents before they are injected into the model prompt.
[DYNAMIC_EXECUTION]: The setup verification script uses dynamic module loading to check the local environment.
Evidence: The file scripts/check-setup.py uses __import__ to programmatically verify the presence of required and optional packages like langchain and langchain_openai. This is a standard and safe implementation for a diagnostic tool.
[EXTERNAL_DOWNLOADS]: The skill documentation encourages the installation of several external packages from the LangChain ecosystem.
Evidence: Files such as references/architecture.md and references/faq-and-troubleshooting.md contain instructions to pip install various well-known packages including langchain, langchain-community, and provider-specific integrations from established organizations.

langchain