The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/await_until.py allows for arbitrary command execution by accepting a shell command string through the --check-cmd argument and running it via subprocess.run. While shlex.split is used, this does not prevent the execution of arbitrary binaries with agent-supplied arguments.
[COMMAND_EXECUTION]: scripts/generate_executor.py dynamically constructs and executes Python code strings (e.g., to verify the executor or inventory scripts) using subprocess.run(['python3', '-c', ...]).
[REMOTE_CODE_EXECUTION]: The core functionality of the skill involves generating and using execute-script.py, which acts as a proxy to resolve and execute other scripts. This pattern, combined with the 'Bootstrap Pattern' described in SKILL.md that executes code from the plugin cache, facilitates the dynamic execution of external or generated code.
[DYNAMIC_EXECUTION]: The scripts/generate_executor.py script performs runtime code generation by populating the templates/execute-script.py.template with absolute paths and version stamps, creating a new executable Python script on the file system.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface in scripts/await_until.py. It ingests output from external commands (--check-cmd), parses it as TOON (Tab-delimited Object Notation), and uses the fields (like status) to determine its own execution flow, which could be manipulated by malicious tool output.

tools-script-executor