Extension Development Guidelines for azd

Overview

This guide provides design guidelines and best practices for developing extensions to the Azure Developer CLI (azd). Following these guidelines ensures that extensions maintain consistency with core azd functionality and provide a seamless user experience.

Design Guidelines for Extensions

1. Command Integration Strategy

• New functionality should extend existing command categories
• Use a verb-first structure, where the primary action (e.g., add, create, delete, list) is the top-level command, and the target entity or context follows as an argument or subcommand.
• Example: azd add <new-resource-type> instead of azd <new-resource-type> add

2. Parameter and Flag Consistency

• Reuse established parameter patterns across new commands
• Maintain consistent naming conventions (e.g., --subscription, --name, --type)
• Provide sensible defaults to reduce cognitive load
• Do not reuse reserved global flag names — see section below

Reserved Global Flags

azd pre-parses a set of global flags from the command line before dispatching to extensions. The extension SDK (NewExtensionRootCommand) also registers these same flags on every extension's root command. Because both azd and the extension parse the same argv, extensions must not register flags that collide with these reserved names.

This is not a new restriction — it has been true since the extension system was designed. The SDK now enforces it at startup via ValidateNoReservedFlagConflicts().

Reserved flag names

Long Name	Short	Purpose
environment	e	Selects the azd environment
cwd	C	Sets the working directory
debug	—	Enables debug logging
no-prompt	—	Non-interactive mode
output	o	Output format (json, table, none)
help	h	Command help (cobra built-in)
docs	—	Opens command documentation
trace-log-file	—	Diagnostics trace file
trace-log-url	—	OpenTelemetry trace endpoint

What this means for extension authors

DO:

• Use any flag name that is not in the table above
• Use any single-letter short flag except e, C, o, h
• Access the environment name via extCtx.Environment (the SDK provides it automatically)
• Ignore reserved flags you don't need — the SDK handles them

DON'T:

• Register --environment or -e on any subcommand (use --env-name or --target-env if you need a second environment reference)
• Register --debug, --cwd, --output, --help, or their short forms for a different purpose
• Assume you can "override" a global flag on a subcommand — azd's pre-parser will consume it first

What happens if you collide

If your extension uses the azd SDK (azdext.Run()), the SDK validates all flags at startup. A collision produces a clear error:

extension defines flags that conflict with reserved azd global flags:
  - command "custom create": flag --endpoint/-e conflicts with reserved global flag --environment
    (short flag -e is reserved by azd for --environment)
Remove or rename these flags to avoid conflicts with azd's global flags.
Reserved flags: environment, cwd, debug, no-prompt, output, help, docs, trace-log-file, trace-log-url

Background

For the full technical specification of how flags flow between azd and extensions, including the pre-parsing pipeline, environment variable propagation, and enforcement implementation, see Extension Flag Architecture Spec.

3. Help and Discoverability

• Integrate new functionality into existing azd help structure
• Provide contextual guidance within established command flows
• Maintain documentation consistency across core and extended features

4. Template and Resource Integration

• Leverage existing template system for new resource types
• Follow established template discovery and management patterns
• Integrate new resources into azd's resource lifecycle management

5. CI/CD and IaC Guidance

• Provide support for GitHub Actions and Azure DevOps
• Consider support for a range of IaC providers (Bicep, Terraform, etc.)

Implementation Benefits

• User Familiarity: Builds on known command patterns and reduces learning curve
• Discoverability: New capabilities are found through existing workflows
• Consistency: Predictable behavior across all azd commands and extensions
• Maintainability: Systematic approach reduces complexity and technical debt
• Extensibility: Clear framework for adding capabilities without breaking existing patterns
• Ecosystem Growth: Provides foundation for third-party extensions and integrations

Future Considerations

This framework enables:

• Advanced workflow automation
• Enhanced developer productivity features
• Consistent user experience across all azd functionality
• Integration of new Azure services and capabilities
• Third-party extension development

Error Handling in Extensions

Extensions communicate with the azd host over gRPC. When an extension returns an error, it must be serialized into an ExtensionError proto message so that the host can display it to users and classify it in telemetry.

Structured Error Types

The azdext package provides structured error types for service, local, and external-tool failures:

• azdext.ServiceError — for HTTP/gRPC service failures (e.g., Azure API returned 429). Fields: Message, ErrorCode, StatusCode, ServiceName, Suggestion.

• azdext.LocalError — for local errors such as validation, auth, config, or internal failures. Fields: Message, Code, Category, CauseTypes, Suggestion.

• azdext.ToolError — for failures from external tools or subprocesses. Fields: Message, Err, ToolName, Kind, ExitCode, Suggestion, Links. Use ToolErrorKindMissing when the tool was not found and ToolErrorKindFailed when it ran and failed.

These types implement Error(). They are detected via errors.As during serialization. CauseTypes is bounded, extension-provided diagnostic input for unexpected local fallbacks; it does not change the selected classification. The host records it only as hashes in error.extension.cause_types, never as plain text in system metadata.

Telemetry Classification

The host classifies extension errors into telemetry codes using the pattern:

Error type	Telemetry code pattern
ServiceError with ErrorCode	ext.service.<errorCode>
ServiceError with StatusCode	ext.service.<serviceName>.<statusCode>
LocalError	ext.<category>.<code>
ToolError	tool.<toolName>.missing or tool.<toolName>.failed
Unclassified	ext.run.failed

Failed extension commands use an ext.run span with the extension ID and version, but no extension.event. Failed lifecycle hooks use the enclosing cmd.* span and include the extension ID, version, and lifecycle event.

For ToolError, the host derives the telemetry name from either POSIX or Windows paths, removes the executable extension, and lowercases the basename. It accepts only 1-64 ASCII characters matching [a-z0-9_-]; invalid or oversized names are recorded as other without changing the displayed error.

Recommended Layering Pattern

Entry-point or orchestration layer: Usually creates structured errors once it can confidently choose the final category, code, and suggestion. This often includes command handlers, top-level actions, or other user-facing coordination code.

Lower-level helpers, parsers, and clients: Usually return plain Go errors with fmt.Errorf("context: %w", err) and let a higher layer classify the failure.

Treat this as guidance, not a strict package boundary. The important part is that structured classification happens in a layer with enough context to produce the right telemetry and a useful suggestion.

Error Chain Precedence

When WrapError serializes an error for gRPC, it checks the chain via errors.As and picks the first match in this order:

1. ServiceError, LocalError, or ToolError (highest priority)
2. azcore.ResponseError (auto-detected Azure SDK errors)
3. gRPC status (host-originated metadata and auth classification)
4. Fallback (unclassified)

Because Go's errors.As walks from outermost to innermost, classifying near the outer orchestration layer naturally produces the intended classification.

Error Code Conventions

Error codes should be:

• Lowercase snake_case (e.g., missing_subscription_id, invalid_agent_manifest)
• Descriptive of the specific failure, not the general category
• Unique within the extension
• Defined as const values (not inline strings) for consistency and grep-ability

Display and UX

When a structured error has a non-empty Suggestion, the azd host displays it as a formatted "ERROR + Suggestion" block. When there is no suggestion, only the error message is shown. Extensions should provide suggestions for user-fixable errors (validation, auth, dependency) and omit them for internal/unexpected errors.

---

For core design principles that apply to all azd functionality, see guiding-principles.md.