The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/run_grok.py executes the grok CLI using the subprocess.run method. It passes the command and its arguments as a list with shell=False (default), which is a secure implementation that prevents shell command injection.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes untrusted data. Ingestion points: The script reads data from args.prompt and user-provided files via args.prompt_files. Boundary markers: Multiple prompt files are separated by a triple-dash delimiter (---). Capability inventory: The skill executes the Grok LLM via subprocess.run, which can perform complex reasoning and tool use. Sanitization: No explicit sanitization or instruction-capping is performed on the ingested content before it is passed to the LLM.

grok-runner