Azure Production Diagnostics

The skill explicitly defaults to read-only diagnostics, and its AKS guidance forbids restarting, deleting, scaling, upgrading, or reconfiguring unless the user explicitly requests remediation. However, other service guides directly provide commands for scaling, changing registry identity, enabling auto-heal, binding certificates, and changing configuration, without uniform step-by-step confirmation, least-privilege guidance, sensitive-log redaction, data-flow disclosure, or rollback plans, so points are deducted.

It provides a systematic flow covering symptoms, resource health, logs, metrics, and recent changes, and tells the agent to stop and ask when AKS identity is missing. However, it depends on Azure CLI, multiple MCP tools, kubectl, Docker, Kudu, and specific log tables; installation, authentication, permission, version-compatibility, and complete tool-failure feedback are underdocumented. Some commands and queries lack sufficient applicability notes. The score stays within the static-review ceiling because execution was not performed and edge coverage is limited.

Triggers cover common Azure production-diagnostics scenarios across App Service, Container Apps, Functions, AKS, Event Hubs, and Service Bus, while AKS inputs and stop conditions are relatively clear. However, the scope is very broad, non-fit cases, input/output formats, service-version differences, Chinese-language interaction, and mainland-China network or regional reachability are not addressed. Some core paths depend on Azure online services and MCP, so points are deducted.

The skill has triggers, a quick flow, service routing, reference directories, and command examples. Metadata includes MIT licensing, Microsoft authorship, and version 1.1.6; the repository also supplies a license, security policy, CI, and work-in-progress maintenance signals. However, there is no changelog, explicit maintenance owner or update path, dependency/install guidance, FAQ, compatibility matrix, or complete limitations disclosure. Progressive disclosure and troubleshooting remain incomplete if referenced documents or MCP tools are unavailable, so points are deducted.

For common Azure diagnostic tasks it supplies reusable CLI, KQL, and MCP templates, evidence ordering, and an output checklist, offering clear value over manual discovery. However, commands use many placeholders, correctness depends on environment, permissions, diagnostic configuration, and tool versions, and remediation suggestions require human review without representative result samples or static verification evidence. The score follows the static ceiling and is reduced for limited completeness and direct usability.

The material contains concrete commands, KQL, input requirements, evidence ordering, and several safety constraints; the repository also includes generic CI, evaluation workflows, and test code. However, the supplied tests do not demonstrate this azure-diagnostics skill's key paths. The workflows mainly cover generic evaluation and infrastructure, with no skill-specific tests, real outputs, independent corroboration, or versioned change evidence, so only limited static auditability is awarded.