The Agent Skills Directory

Command Injection Consideration: The discover_and_rank.sh script interpolates shell variables directly into a Python command string, which could be exploited if input parameters are not properly sanitized. * Evidence: The script uses python3 -c "... print(f'Model: {\"${MODEL_NAME}\"} ...') " where ${MODEL_NAME} is a shell variable expanded directly inside the command-line argument.
Azure API Interaction: The skill uses az rest to interact with Azure Management APIs to retrieve resource and capacity data. * Evidence: Requests are made to https://management.azure.com/subscriptions/ endpoints to gather model capacities, account lists, and usage statistics.
Local Automation Scripts: The skill relies on pre-built PowerShell and Bash scripts to automate complex cloud management queries. * Context: The workflow assumes the presence of authenticated az CLI, jq, and python3 in the local execution environment.
Indirect Prompt Injection Surface: The skill processes untrusted user input and interpolates it into shell commands and scripts without explicit boundary markers or sanitization. * Evidence: User-provided model names and versions are passed as positional arguments to shell scripts that execute command-line tools and API requests.

capacity