The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to interact with the external qwen CLI tool to facilitate model requests, delegating the primary task to an external process.
[COMMAND_EXECUTION]: The script scripts/run_minimax.py performs dynamic loading by modifying the Python search path at runtime to import modules from a computed relative path.
Evidence: sys.path.insert(0, str(QWEN_RUNNER_DIR)) where QWEN_RUNNER_DIR is resolved to ../../qwen-runner/scripts relative to the file.
[PROMPT_INJECTION]: The skill exposes an attack surface for indirect prompt injection by accepting and processing untrusted user data as model prompts without evident sanitization.
Ingestion points: User prompts provided as command-line arguments to scripts/run_minimax.py or via the --prompt-file option.
Boundary markers: No delimiters or explicit instructions to ignore embedded instructions are implemented in the wrapper script.
Capability inventory: The skill has the capability to execute shell commands via the qwen CLI.
Sanitization: The input is passed directly to the execution logic without filtering or escaping.

minimax-runner