The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/classify_error.py uses importlib.import_module() to dynamically load Python modules specified through command-line arguments, which allows for the execution of side effects from arbitrary local modules.
[DYNAMIC_EXECUTION]: The script scripts/wrap_with_retry.py generates executable Python code at runtime using string templates based on user-provided CLI flags.
[INDIRECT_PROMPT_INJECTION]: The error-handling patterns demonstrated in assets/examples/retry-example/python/graph.py and assets/examples/human-loop-example/python/graph.py establish an indirect prompt injection surface. Ingestion points: Raw exception messages from external tools and APIs are captured and stored in the graph state. Boundary markers: No delimiters or safety instructions are used when updating the state with error content. Capability inventory: The agent node reads the untrusted error strings from the state to determine routing logic via Command objects. Sanitization: No filtering or escaping is applied to the exception strings before they are added to the conversation state.

langgraph-error-handling