The Agent Skills Directory

Command Execution: The skill utilizes shell commands (bash/PowerShell) and Node.js scripts to interact with development tools like the Power Platform CLI (pac) and Azure CLI (az). These operations are used for infrastructure tasks such as authentication, environment discovery, and resource deployment.
Indirect Prompt Injection Surface: During the edit flow, the skill ingests and processes existing page source code, comments, and metadata. To mitigate the risk of embedded malicious instructions, the instructions define a specific "Edit data boundary" that treats all such ingested data as untrusted and limits its influence over the agent's actions.
External Communications: The skill communicates with Power Platform environment URLs (e.g., crm.dynamics.com) to perform authenticated operations. These interactions are necessary for its primary function of managing and deploying model-driven pages.
Injection Prevention Best Practices: The skill follows robust security patterns for handling dynamic text. Instead of passing arbitrary user input directly as shell arguments, it writes content such as prompts and display names to temporary files, which are then referenced by the tools to avoid command injection vulnerabilities.
Sensitive Data Management: Implementation includes logic to redact secrets, connection IDs, and sensitive tokens from logs and console output, ensuring that credentials are not inadvertently exposed during the workflow.

genpage