The Agent Skills Directory

[COMMAND_EXECUTION]: The project initialization scripts scripts/init_langgraph_project.py and scripts/init_langgraph_project.js take a user-provided project name to generate the directory structure. These scripts use the input directly in filesystem path construction without sanitizing for path traversal characters like ../, which could lead to directory creation in unintended locations if executed with malicious arguments.
[INDIRECT_PROMPT_INJECTION]: The skill generates boilerplate code for LangGraph agents that ingest and process message data. The templates provided in the initialization scripts do not include default input validation or boundary markers, creating an architectural surface for indirect prompt injection if the resulting agent processes untrusted data without further customization.
[DYNAMIC_EXECUTION]: The documentation in references/langgraph-json-schema.md references the checkpointer configuration which supports a pickle_fallback option. While this is a standard framework feature, using Python's pickle module for state persistence can pose a risk of arbitrary code execution if the persisted state is tampered with or originates from an untrusted source.

langgraph-project-setup