CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

aks-troubleshooting

Debug live Azure Kubernetes Service (AKS) incidents with a read-only, evidence-first investigation. WHEN: pod crashes or Pending, CrashLoopBackOff, OOMKilled, ImagePullBackOff, node NotReady, DNS or ingress failure, connectivity timeout, network policy, SNAT exhaustion, node-pool scaling blocked by QuotaExceeded or InsufficientVCPUQuota, upgrade stuck, spot or zone disruption, a bare VMExtensionProvisioningError or AllocationFailed wrapper, an uncataloged capacity symptom, or 'investigate my AKS cluster'. DO NOT USE FOR: packet capture (use aks-network-capture); GPU or model-serving issues (use aks-gpu-inference); cluster creation or provisioning (use azure-kubernetes); cost (use cost-analysis from the optional azure-cost plugin); pod rightsizing (use azure-kubernetes); a fully qualified documented AKS signature with every required nested qualifier (use aks-known-issues); standalone failures on non-AKS Azure resources (use azure-diagnostics). Unqualified errors and open-ended incidents stay here only for AKS.

Quality

88%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Low

Low-risk findings worth noting

SKILL.md
Quality
Evals
Security

Quality

Content

78%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A well-crafted troubleshooting body: lean, AKS-specific gotchas instead of generic Kubernetes education, copy-paste commands, and a genuinely convergent diagnostic methodology with pre-flight and re-test checkpoints. Its main defects are the distributed (rather than consolidated) workflow presentation and a bundle-integrity problem where most of the symptom routing table links to topic files that are missing from the bundle as provided.

Suggestions

Ship (or restore) the eight topic files referenced in the routing table (pod-failures.md, node-issues.md, networking.md, load-balancer-and-ingress.md, network-policy.md, upgrade-operations.md, spot-and-zone-issues.md, general-diagnostics.md) or repoint those rows to existing files such as references/symptom-map.md so no navigation link is broken.

Consolidate the Operating rules, Evidence order, and Host capability gate sections into one ordered end-to-end procedure (bind -> route -> collect -> converge -> report) with the validation checkpoints embedded at each step.

Deduplicate the symptom-map description (currently in the route table, the following paragraph, and Log discipline) and trim the overlap between Tool preference and Host capability gate.

DimensionReasoningScore

Conciseness

The body is dense with AKS-specific, non-generic knowledge (SNAT exhaustion past a few hundred nodes, VM extension error codes 50/51/52, API-server IP change after stop/start, default max-surge of one node, kubectl version-skew rule) and avoids explaining concepts Claude already knows. Not 5 because a few sections could be tightened: the "Host capability gate" rule overlaps the "Tool preference" rule for several lines, and the symptom-map description is repeated in three places (route table row, dedicated paragraph, Scripts/Log discipline sections).

4 / 5

Actionability

Guidance is copy-paste executable throughout: exact commands with their queries ("az aks show -o json --query networkProfile.networkPlugin", "kubectl get pdb -A", "az network lb probe list", "kubectl auth can-i", "date -u"), concrete script invocations with argument lists ("scripts/cluster-snapshot.sh <resource-group> <cluster> <kube-context>"), and a pointer to a 17-section symptom map of exact commands in order. The common cases are covered with specific, runnable commands; this matches the fully executable anchor.

5 / 5

Workflow Clarity

The diagnostic sequence is clearly defined — bind subscription/cluster/context first, let the symptom select the first decisive read (workload-local vs control-plane branches), follow the causal branch, converge on 2-4 hypotheses with confirming/falsifying signals, stop at one supported cause or report the gap — with real validation checkpoints (host capability pre-check, script-side context-endpoint verification, re-test of the original signal after any remediation). Not 5 because the end-to-end procedure must be assembled from six separate prose sections rather than one coherent ordered walkthrough, and there are no checklists for the report/escalation stage.

4 / 5

Progressive Disclosure

The in-body structure itself is strong — a symptom routing table, one-level-deep clearly-signaled references, and pointers to scripts and a report template — but scored against the actual bundle, 8 of the routing-table targets (general-diagnostics.md, pod-failures.md, node-issues.md, networking.md, load-balancer-and-ingress.md, network-policy.md, upgrade-operations.md, spot-and-zone-issues.md) do not exist anywhere in the provided bundle, and references/symptom-map.md itself links to ../node-issues.md. More than half of the primary navigation routes are broken, which prevents the clear navigation the top anchors require; the well-organized existing structure is the only reason this is not 2.

3 / 5

Total

16

/

20

Passed

Description

95%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

An excellent description: third-person, explicit WHAT/WHEN structure, comprehensive natural trigger terms including exact Azure error codes, and explicit de-confliction against six sibling skills. The only (minor) limitation is that the capability is expressed through one action verb plus scope enumeration rather than several distinct concrete actions.

DimensionReasoningScore

Specificity

"Debug live Azure Kubernetes Service (AKS) incidents with a read-only, evidence-first investigation" names the domain and a concrete investigative mode, and the WHEN clause enumerates specific failure classes (CrashLoopBackOff, OOMKilled, ImagePullBackOff, node NotReady, SNAT exhaustion, QuotaExceeded/InsufficientVCPUQuota, upgrade stuck). Not 5 because it relies on a single action verb (debug/investigate) rather than naming multiple distinct operations (e.g., gather evidence, isolate root cause, produce an incident report); not 3 because coverage of the domain's concrete scenarios is far more comprehensive than 1-2 actions.

4 / 5

Completeness

Both questions are answered explicitly: WHAT — "Debug live AKS incidents with a read-only, evidence-first investigation"; WHEN — a literal "WHEN:" clause with concrete trigger phrases plus a quoted natural-language trigger ("investigate my AKS cluster"), and additionally a DO NOT USE FOR clause. This clearly matches the anchor for explicit what and when with concrete triggers; nothing is left implicit.

5 / 5

Trigger Term Quality

The description includes the exact natural terms an AKS operator would say — "pod crashes", "CrashLoopBackOff", "OOMKilled", "node NotReady", "DNS or ingress failure", "upgrade stuck", "investigate my AKS cluster" — plus precise error codes (QuotaExceeded, InsufficientVCPUQuota, VMExtensionProvisioningError, AllocationFailed), the AKS equivalent of file extensions in the anchor. This comprehensively covers natural phrasing, synonyms, and exact error identifiers; nothing common is missing.

5 / 5

Distinctiveness Conflict Risk

The DO NOT USE FOR clause disambiguates against every neighboring skill by name (aks-network-capture, aks-gpu-inference, azure-kubernetes, cost-analysis, aks-known-issues, azure-diagnostics) with precise boundary conditions ("a fully qualified documented AKS signature with every required nested qualifier", "Unqualified errors and open-ended incidents stay here only for AKS"). The niche (live AKS Day-2 incident investigation) is clear and conflict risk is minimal.

5 / 5

Total

19

/

20

Passed

Validation

93%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 15 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

relative_links

Relative link issues: 9 missing

Warning

Total

15

/

16

Passed

Repository

microsoft/GitHub-Copilot-for-Azure

Path

plugins/aks-skills/skills/aks-troubleshooting/SKILL.md

Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.