The Agent Skills Directory

[SAFE]: The skill consists of instructional material and standard Python snippets for the LangChain framework. All code examples provided are for educational purposes and do not involve unauthorized system access or command execution.
[INDIRECT_PROMPT_INJECTION]: The architectural patterns for agents and retrieval-augmented generation (RAG) pipelines describe how to process external data sources (e.g., TextLoader for documents.txt) and user input in chains. These templates demonstrate a potential attack surface for indirect prompt injection, which is inherent to the framework's use case and not a malicious defect of the skill itself.
Ingestion points: TextLoader for processing external files and input variables in the SequentialChain examples.
Boundary markers: No specific delimiters or prompt-injection guardrails are implemented in the provided prompt templates.
Capability inventory: The documentation provides examples for tools that can search databases or send emails.
Sanitization: No input validation or escaping logic is included in the architectural snippets.
[EXTERNAL_DOWNLOADS]: The skill mentions external supporting files (e.g., assets/agent-template.py) but explicitly advises the agent to verify these files locally and avoid fabricating them or downloading from unverified sources.

langchain-architecture