The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes powerful infrastructure management commands, including Azure CLI (az deployment) and Terraform (terraform apply).
Evidence: Instructions in deployment.md and workflow.md detail the execution of these commands for resource provisioning.
Mitigation: The risk is significantly reduced by mandatory human-in-the-loop gates requiring user approval of the infrastructure-plan.json before any code is generated or executed, and explicit confirmation for destructive actions.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it processes untrusted data from the user's project repository and external documentation.
Ingestion points: Scans local files like package.json, requirements.txt, and Dockerfile (research.md), and fetches documentation from Microsoft Learn URLs.
Boundary markers: Employs sub-agents to summarize external content and requires the final plan to be written to a structured JSON file for user review.
Capability inventory: The agent can execute administrative CLI commands and write to the file system.
Sanitization: Incorporates syntax validation (az bicep build, terraform validate) but relies primarily on the user for final verification of the plan logic.
[EXTERNAL_DOWNLOADS]: The skill fetches resource schemas, naming rules, and architectural guidance from official Microsoft domains.
Evidence: Calls microsoft_docs_fetch for URLs like learn.microsoft.com/azure/templates/ as seen in the resource category files (e.g., resources/compute-infra.md).
Context: These references target a well-known service to ensure the generated infrastructure adheres to current platform specifications.

azure-enterprise-infra-planner