The Agent Skills Directory

Command Execution: The skill interacts with the Azure CLI (az) to authenticate and manage Azure resources. These interactions are performed using subprocess calls with arguments that are validated against expected patterns (e.g., UUIDs and resource names) to prevent unintended execution.
Sensitive Data Handling: A dedicated "reject secrets" mechanism is implemented to prevent sensitive information like API keys or connection strings from being included in plans or outputs. Additionally, access to common credential file paths (such as ~/.ssh, .aws, and .env) is explicitly blocked by the skill's logic.
Isolated Document Analysis: When analyzing local documents to determine their structure, the skill utilizes a separate worker process with operating-system-level resource constraints (memory and CPU limits) and Python audit hooks to prevent unauthorized system operations, such as network access or file writes, during parsing.
Secure Receipt Management: The skill maintains private journals and receipts for operations, which are stored in directories with restrictive permissions (e.g., 0700 for directories and 0600 for files) and verified for owner identity and stability to ensure evidence integrity.

foundry-iq