The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The scripts/migrate_state.py utility allows for the execution of arbitrary Python code.
Evidence: The function load_migration_script in scripts/migrate_state.py uses importlib.util.spec_from_file_location and spec.loader.exec_module(module) to load and run a file path provided by the user at the command line.
Risk: An attacker who can influence the command-line arguments of this script (e.g., through prompt injection or malicious file placement) can achieve full Remote Code Execution (RCE) on the host system.
[DATA_EXFILTRATION] (MEDIUM): The migration script accesses and logs state data from SQLite databases.
Evidence: The migrate_all and _print_diff methods read checkpoint data and print differences to standard output.
Risk: Agent states often contain sensitive information, including chat history, user data, or transient secrets. This information is exposed during the migration process and could be logged or intercepted.
[REMOTE_CODE_EXECUTION] (HIGH): While the script targets local files, the pattern of loading and executing external logic is a primary vector for RCE in agentic environments.
Evidence: Combined use of importlib and exec_module on a user-defined path.

langgraph-state-management