The Agent Skills Directory

Indirect Prompt Injection Surface: The skill analyzes repository files such as package.json, docker-compose.yml, and .env.example to determine deployment requirements. Maliciously crafted project files could potentially influence the generated infrastructure code or deployment commands.
Ingestion points: Files read in scaffold/references/subagent-iac-gen.md and deploy/references/database-post-deploy.md to map dependencies and environment variables.
Boundary markers: The skill utilizes specific template delimiters (<<<TEMPLATE_START>>>) for sub-agent communication and session-specific identifiers.
Capability inventory: The orchestrator spawns subprocesses for Azure CLI (az), Static Web Apps CLI (swa), and PowerShell/Bash scripts to manage infrastructure and code deployment.
Sanitization: The skill implements syntax validation (az bicep build), adversarial self-review layers, and automated conformance scripts (scaffold-conformance.ps1) to verify generated code before execution.
Dynamic Infrastructure Generation: The skill dynamically generates Bicep and Terraform templates at runtime based on the detected application architecture. While this enables tailored deployments, it involves the assembly and execution of code generated from local project metadata.
External Tools from Trusted Sources: The skill utilizes standard package managers to download and execute official Microsoft utilities, such as the @azure/static-web-apps-cli. These are official vendor resources and are used for their intended purpose in the deployment workflow.
Command Execution for Azure Management: The skill performs a wide range of Azure management operations via the command line. To mitigate risks, it includes an extensive list of 'Blocked Patterns' that prevent the execution of destructive or unsafe commands during automated workflows.

azure-app-onboard