The Agent Skills Directory

[SAFE]: The skill functions as a technical guide for the LangGraph library, providing standard implementation patterns for agentic workflows.
[SAFE]: Includes explicit security instructions in 'references/detailed-guide.md', warning against the use of unrestricted evaluation and recommending bounded, safe implementations for math tools.
[SAFE]: Promotes secure production practices, such as isolating persistence databases for different users and using interrupt mechanisms to require human approval before executing sensitive tool actions.
[INDIRECT_PROMPT_INJECTION]: The skill demonstrates patterns for agents that process user-controlled input (messages) within an LLM context, which constitutes a standard attack surface for indirect prompt injection. The documentation mitigates this by instructing the user on architectural guardrails.
Ingestion points: 'AgentState' dictionary in 'references/detailed-guide.md' which aggregates user messages.
Boundary markers: None defined within the code snippets, but the text emphasizes that tool usage must be 'bounded by the application'.
Capability inventory: Tool definitions for 'search' and 'calculator' in 'references/detailed-guide.md'.
Sanitization: Recommends using a 'safe_math_evaluator' instead of 'eval()' and mentions redacting sensitive state content.

langgraph