The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/run_command.py executes arbitrary shell commands provided via the --command argument using subprocess.run. While it utilizes shlex.split to handle argument parsing, the script essentially acts as a wrapper for executing system-level commands determined by the agent or user input.- [COMMAND_EXECUTION]: The script scripts/write_outputs.py performs dynamic loading and execution of a Python module from a relative path (../../../shared/scripts/write_run_bundle.py). This dynamic execution pattern relies on the integrity of the target file system and files located outside the skill's own directory.- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it processes externally provided reproduction goals and commands and executes them. \n
Ingestion points: Commands and reproduction goals passed to scripts/run_command.py.\n
Boundary markers: The instructions lack explicit delimiters to separate untrusted data from the execution prompt.\n
Capability inventory: subprocess.run in scripts/run_command.py allows for system-level actions.\n
Sanitization: Arguments are split using shlex.split, but the commands themselves are not validated against a whitelist.

minimal-run-and-audit