fix: unify split toolbox diagnostics

[Hui Miao (huimiu)]

Why this is needed

Projects with a unified azure.yaml can declare toolboxes as standalone host: azure.ai.toolbox services. Agent diagnostics previously read only legacy manifest toolboxes, so split toolboxes were left out of state and their managed endpoint variables could be treated as manual configuration. Next-step commands could then be wrong, and Doctor could skip the check or report a misleading result.

What changed

• Discover standalone toolbox services and merge them with the existing toolbox state using the canonical endpoint key.
• Probe each collected toolbox endpoint directly. Keep missing managed endpoints separate from manual variables, and preserve endpoint read errors.
• Show azd deploy guidance for split toolbox endpoints while preserving provisioning precedence and avoiding duplicate deploy suggestions.
• Make Doctor use the assembled toolbox state and reuse one state snapshot for its checks and trailing guidance.
• Add regression coverage for split discovery, endpoint classification, source-aware guidance, mixed sources, and state-cache reuse.

Scope

This PR covers only the toolbox part of #8710. Model and connection migration, runtime behavior, and provisioning implementation are unchanged. Legacy toolbox migration remains a separate follow-up.

Fixes #9566
Related to #8710

[github-actions]

📋 Prioritization Note

Thanks for the contribution! The linked issue isn't in the current milestone yet.
Thank you for logging this issue; our team is reviewing it. If you need urgent prioritization, tag Rick Winter (@RickWinter) and Kristen Womack (@kristenwomack) to let us know.

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).
20 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[Copilot]

Pull request overview

Unifies toolbox diagnostics for standalone azure.ai.toolbox services across next-step and Doctor flows.

Changes:

• Discovers and probes split toolbox endpoints.
• Adds source-aware deploy/provision guidance.
• Shares one assembled state snapshot across Doctor checks.

Reviewed changes

Copilot reviewed 15 out of 15 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
nextstep/types.go	Adds toolbox source and probe state.
nextstep/state.go	Probes canonical toolbox endpoints.
nextstep/state_test.go	Tests split toolbox state assembly.
nextstep/resolver.go	Adds source-aware remediation.
nextstep/resolver_test.go	Tests split toolbox guidance.
nextstep/manifest.go	Discovers standalone toolbox services.
doctor/state_cache.go	Caches assembled diagnostic state.
doctor/state_cache_test.go	Verifies cache reuse.
doctor/checks_toolboxes.go	Uses assembled toolbox probe results.
doctor/checks_toolboxes_test.go	Tests toolbox classifications.
doctor/checks_manual_env.go	Reuses cached state.
doctor/checks_local.go	Shares the state cache.
doctor/checks_local_test.go	Updates toolbox check expectations.
doctor/checks_connections.go	Reuses cached state.
doctor.go	Shares state with trailing guidance.

---

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Pull request overview

Copilot reviewed 15 out of 15 changed files in this pull request and generated no new comments.

[Copilot]

Pull request overview

Copilot reviewed 15 out of 15 changed files in this pull request and generated no new comments.

[Copilot]

Pull request overview

Copilot reviewed 15 out of 15 changed files in this pull request and generated no new comments.

Suppressed comments (2)

cli/azd/extensions/azure.ai.agents/internal/cmd/nextstep/types.go:162

• azd-code-reviewer: The following paragraph still says all three flags are false when a manifest is absent or malformed, which now contradicts split-toolbox discovery: HasToolboxes can be true without any manifest. Update the field documentation so callers do not infer the old manifest-only contract.

	// HasModels, HasToolboxes, and HasConnections are aggregate flags.
	// They describe resources. Models and connections come from
	// agent manifests. Toolboxes include split services and manifest
	// resources. Doctor checks skip when no matching resource exists,
	// while resolvers can tailor remediation.

cli/azd/extensions/azure.ai.agents/internal/cmd/doctor/state_cache.go:32

• azd-code-reviewer: This fallback does not actually cache across calls because Dependencies is received by value. When StateCache is nil, every call creates a new cache and reruns state assembly, contrary to this method's contract. Persist the cache by using a pointer receiver.

func (deps Dependencies) AssembleAgentState(ctx context.Context) (*nextstep.State, []error) {
	cache := deps.StateCache
	if cache == nil {
		cache = NewStateCache()
	}

[Travis Angevine (trangevi)]

[azd-code-reviewer] This collects every azure.ai.toolbox service in the project, even when no agent references it through uses (or when its deployment condition disables it). An unrelated toolbox with no endpoint then populates MissingToolboxEndpoints, so ResolveAfterInit blocks local-run guidance and tells the user to deploy a resource the agent does not need. Filter split toolboxes to the dependencies of the collected agent services, and honor deployment conditions or avoid treating disabled services as missing.