Skill Instructions

Goal

Generate a new, human-maintainable set of .bicep files plus a .bicepparam file that can redeploy the current live state of a target Azure resource group as a no-op (i.e., deploying should produce no changes).

Guiding Principles

• •Optimize for maintainability and clarity: split into sensible modules, use descriptive names, and factor repeated patterns.
• •Prefer templates that look like they were written by a human, not a direct export dump.
• •If there is a trade-off between “prettier” and “no-op correctness,” correctness wins.

Inputs

Ensure you have the subscriptionId and resourceGroup of the resource group to export from the user before starting.

Required Tools / Commands

Strongly prefer the following tools/commands as part of the workflow, instead of finding other tools or CLI commands to call:

• •get_bicep_best_practices MCP tool: learn current best practices before refactoring.
• •get_bicep_file_diagnostics MCP tool: compile/validate Bicep and address errors/warnings during edits.
• •get_deployment_snapshot MCP tool: predict deployment output; take “before” and “after” snapshots and compare.
• •what_if_deployment MCP tool: run a live diff against Azure using a .bicepparam file.
• •az group export ... CLI command: export the live resource group to bootstrap the initial template.

Workflow

1) Bootstrap from Live State (Export)

• •Run az group export ... against the target resource group to get an initial baseline. This will give you a very rough machine-generated output.
• •Use the --export-format bicep and --output tsv arguments to obtain a .bicep file output.
• •Use the --skip-all-params argument to avoid generating parameter statements.
• •Only pipe stdout, don't use 2>&1 to pipe stderr to file.
• •Treat export output as input material only; plan to refactor heavily for readability and long-term maintenance.
• •Generate an empty .bicepparam file, pointing to the exported .bicep file - for example, assuming it is named .main.bicep:
  bicep

  using 'main.bicep'
  

• •Clearly differentiate the fact that these files are raw exports by naming them exported_raw.bicepparam and exported_raw.bicep

2) Capture snapshot file

• •Capture a snapshot for the “before” state (initial exported structure + parameters) using the get_deployment_snapshot MCP tool.
• •Ensure you pass it all the inputs (subscription id, resource group name) that you have values for.
• •Save the MCP tool output to exported_raw.snapshot.json, bearing in mind it is a large file.

3) Refactor for maintainability

Refactor the exported output into a clean structure:

• •Do not modify the raw exported files, create a new file structure instead.
• •Use a main entrypoint (for example main.bicep) and split resources into modules by domain (for example: network, compute, storage, identity, monitoring).
• •Use parameters and variables judiciously: avoid abstraction for its own sake; prefer clarity. Examples where you might want to use parameters are valuse that the user may want to vary between deploymens to different environments - e.g. locations, resource base names, subscription Ids, application Ids.
• •Declare parameter values in the .bicepparam file, rather than using default values in the .bicep file parameter declarations.
• •Prefer stable, explicit resource names and scopes.
• •Preserve drift-sensitive configuration needed for a no-op (examples: identity blocks, SKU/tier, role assignments, locks, private endpoints/DNS, diagnostics settings).
• •Use snapshots when embarking on complex refactors to verify that the before+after produce the same desired state.

4) Keep the Template Healthy (Diagnostics)

• •After each significant change, run get_bicep_file_diagnostics on the edited .bicep files.
• •Fix compilation errors immediately.
• •Review warnings and fix anything that could affect correctness, deployment behavior, or maintainability.

5) Predict Deployment Output (Snapshots)

This provides confidence that the code has been succesfully refactored without altering the outcome of the deployment.

• •Capture a new snapshot of the “after” state (refactored modules + parameters) using the get_deployment_snapshot MCP tool.
• •Ensure you pass it all the inputs (subscription id, resource group name) that you have values for.
• •Save the MCP tool output to main.snapshot.json, bearing in mind it is a large file.
• •Diff the new snapshot against the original snapshot, to ensure that all of the refactors have succesfully preserved semantics.
• •If there are any semantic differences, iterate until snapshots are equivalent (or any differences are explicitly intentional and justified).

6) Validate Against Live Azure (What-If)

• •Once snapshots look equivalent, use the what_if_deployment MCP tool, using the .bicepparam file.
• •Review the full What-If output.
• •The target outcome is a no-op: no creates, no deletes, no updates.
• •If What-If reports changes, adjust templates/parameters until What-If is clean.

7) Cleanup

• •Remove the temporary files (exported raw .bicep and .bicepparam files, and snapshots)

Acceptance Criteria

• •A coherent set of .bicep files + a .bicepparam file exists for the target resource group.
• •The Bicep compiles cleanly (diagnostics show no errors; warnings are understood/acceptable).
• •“Before vs after” snapshots are equivalent (or differences are explicitly intentional and justified).
• •what_if_deployment indicates the deployment would have no impact on live resources.