The Agent Skills Directory

[PERSISTENCE]: The scripts/setup_proxy.sh script modifies the user's ~/.bashrc file to persist proxy environment variables (HTTP_PROXY, HTTPS_PROXY, etc.) across shell sessions. While intended for configuration management, modifying shell initialization files is a recognized persistence mechanism.
[COMMAND_EXECUTION]: Both scripts/run_ms_model_download.sh and scripts/run_ms_datasets_download.sh use the eval command to execute dynamically constructed shell strings. This pattern is used to run the modelscope download command with variable arguments such as model IDs and local directories.
[COMMAND_EXECUTION]: The scripts/run_network_diagnose.sh and scripts/run_preflight_check.sh scripts utilize curl -k (or --insecure), which bypasses SSL certificate verification when testing connectivity to ModelScope endpoints.
[EXTERNAL_DOWNLOADS]: The skill is designed to download large-scale data and code (models and datasets) from modelscope.cn, a well-known machine learning platform. It also configures pip to use official third-party mirrors from Tsinghua University and Alibaba for package installation.
[INDIRECT_PROMPT_INJECTION]: The skill provides an interface for the agent to ingest external, untrusted content from the ModelScope repository into the local execution environment.
Ingestion points: Data is fetched via modelscope download commands in scripts/run_ms_model_download.sh and scripts/run_ms_datasets_download.sh.
Boundary markers: None identified; downloaded content is not isolated from subsequent agent context.
Capability inventory: The skill possesses capabilities for filesystem writes (mkdir), integrity checking (sha256sum), and network operations (curl, ping, nc).
Sanitization: None identified; the skill performs integrity verification (SHA256) but does not sanitize the contents of the downloaded models or datasets before the agent interacts with them.

modelscope-cli