The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes external data which could potentially contain malicious instructions.
Ingestion points: Site configuration is read from powerpages.config.json (Phase 2.1), and user input is accepted for site subdomains (Phase 2.2).
Boundary markers: The skill implements an explicit confirmation gate (Phase 3) requiring user approval before any mutating actions. It also validates custom subdomains for alphanumeric characters (Phase 2.2).
Capability inventory: The skill executes several shell commands (pac, az, node) and performs file system writes (node -e in Phase 5.1b).
Sanitization: Subdomain inputs are validated against a specific character set (lowercase alphanumeric with hyphens).
Dynamic Script Execution: In Phase 5.1b, the skill uses node -e to execute dynamically generated JavaScript snippets for directory creation and JSON metadata writing. While these are used for routine local administrative tasks, the generation of executable strings from variable data is a pattern that warrants standard review.
Administrative Command Execution: The skill relies on official Microsoft command-line utilities, including the Power Platform CLI (pac) and Azure CLI (az), to perform site provisioning. It provides instructions to install the Microsoft.PowerApps.CLI.Tool from official registries if it is missing, which is a standard procedure for this vendor's tools.
User Confirmation Gate: A beneficial safety feature is the mandatory confirmation step in Phase 3. The agent is instructed to present all collected activation parameters to the user for explicit approval before proceeding with the API call, ensuring transparency and user control over the provisioning process.

activate-site