The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes bash blocks in Step 3 and Step 4 that interpolate user-provided arguments directly into shell commands. This lacks sanitization, enabling arbitrary command execution if a user provides malicious input containing shell control characters such as semicolons, ampersands, or backticks.
[EXTERNAL_DOWNLOADS]: Retrieves data from https://ollama.com/search?c=cloud using curl to dynamically generate a list of available models. This targets a well-known service and the request is limited to model discovery.
[COMMAND_EXECUTION]: Uses dynamic context injection syntax to execute shell commands at skill load time to check for the presence of the ollama binary and read existing configuration from the ~/.claude/pza-ollama-model file.

ollama-setup