The Agent Skills Directory

[PERSISTENCE_MECHANISMS]: The script scripts/setup_proxy.sh modifies the user's ~/.bashrc file to persist proxy environment variables across sessions. While intended for configuration, modifying shell startup profiles is a persistence technique that affects the entire user environment.
[COMMAND_EXECUTION]: The scripts run_ms_datasets_download.sh and run_ms_model_download.sh utilize eval to execute constructed shell commands. Using eval with variables can lead to command injection vulnerabilities if those variables (such as model or dataset identifiers) are sourced from untrusted inputs or contain malicious shell metacharacters.
[EXTERNAL_DOWNLOADS]: The skill facilitates downloading large files from modelscope.cn and instructs the user to install the modelscope Python package via pip. These interactions target well-known and established service providers.
[CREDENTIALS_UNSAFE]: The scripts/setup_proxy.sh utility prompts users for proxy configurations that may include plaintext credentials. These secrets are subsequently stored in plaintext within the ~/.bashrc file, increasing the risk of credential exposure.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process model and dataset metadata from the ModelScope platform which could potentially contain malicious instructions.
Ingestion points: Model and dataset identifiers processed in run_ms_model_download.sh and run_ms_datasets_download.sh.
Boundary markers: Absent; data is interpolated directly into command strings.
Capability inventory: File system modification (mkdir), network access (curl), and shell command execution (eval).
Sanitization: The skill does not implement validation or escaping for external identifiers before they are executed via the shell.

modelscope-cli