flowing
Warn
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill uses Python's 'pickle' module to implement its durable journal feature, which caches task results to a file. Python's documentation warns that the pickle module is not secure against erroneous or maliciously constructed data. Evidence: The '_load_journal' method in 'scripts/flowing.py' calls 'pickle.loads()' on data read from the user-provided 'journal_path'. A poisoned journal file could be used to execute arbitrary commands when the workflow is resumed or re-run.
- [COMMAND_EXECUTION]: The skill is designed to execute arbitrary Python functions defined by the agent as tasks. Evidence: The 'TaskDef' class and the '_run_step' function in 'scripts/flowing.py' facilitate the execution of user-supplied Python callables. While this is the intended functionality of a DAG runner, it provides a high-privilege execution environment that requires careful oversight.
Audit Metadata