The Agent Skills Directory

[DYNAMIC_EXECUTION]: The inspect_routes.py script utilizes importlib.import_module to dynamically load a specified module for FastAPI route inspection. Similarly, check_component_skeleton.py uses importlib.util.spec_from_file_location and exec_module to perform runtime validation of Langflow components. These are functional requirements for providing developer introspection tools for the platform.
[INDIRECT_PROMPT_INJECTION]: As an agent platform, Langflow naturally processes untrusted workflow definitions (JSON) and user inputs. The skill addresses this vulnerability surface through several mechanisms:
Ingestion points: lfx run, lfx serve, and the validate_flow_json.py helper process external JSON workflow files.
Boundary markers: The documentation explicitly instructs users not to execute untrusted flows without inspection and provides a normalize_flow_file.py script to strip secrets.
Capability inventory: The Langflow runtime (executed via lfx or langflow) has broad capabilities including network access, file system operations, and subprocess execution depending on the components used in a flow.
Sanitization: The normalize_flow_file.py script provides deterministic JSON output and secret stripping (clearing template values marked as passwords or database-backed), while validate_flow_json.py performs static topology checks to prevent structural errors.
[DATA_EXFILTRATION]: The skill includes robust secret management practices. The check_env_vars.py script automatically redacts values containing sensitive patterns such as 'KEY', 'TOKEN', 'SECRET', and 'PASSWORD'. Additionally, the normalize_flow_file.py tool defaults to stripping credentials from flow exports to ensure that workflow definitions are safe for version control.
[COMMAND_EXECUTION]: The executor-cli documentation describes the use of lfx run and lfx serve to execute workflows. This includes support for Python graph scripts, with explicit warnings to the user that such scripts should only be run if trusted, as import-time code will execute.

langflow