# Triage Kubernetes Incidents From Evidence

Kubernetes incident reports often lack enough evidence to identify the next useful check. This skill separates facts from hypotheses and recommends focused checks without claiming cluster access.

## Install

```bash
npx skillstore add ghostwritten/kubernetes-triage-expert
```

## Metadata

- Status: approved
- Slug: ghostwritten-kubernetes-triage-expert
- Skillstore revision: r1
- Version status: missing
- Tree hash: 5522567d0d1b83d7beb126de6054b3ae0c0a25d973c85e6e984f1e1ad99117f1
- Author: ghostwritten
- GitHub username: ghostwritten
- Repository: https://github.com/ghostwritten/kubetriage/tree/002c04c82c40f7765c4ccc9f300af68203ad7c9a/
- Ref: 9ab00e3831591d1ebb99aaab4d585ca2af794805
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, env\_access, filesystem
- Quality score: 60
- Public page: https://skillstore.pages.dev/skills/ghostwritten-kubernetes-triage-expert
- Manifest: https://skillstore.pages.dev/api/skills/ghostwritten-kubernetes-triage-expert/manifest

## Capabilities

- Normalize incident reports into affected objects, symptoms, timing, impact, recent changes, and strongest evidence.
- Classify startup, scheduling, networking, storage, resource, and rollout failures from supplied evidence.
- Rank up to three hypotheses using evidence fit, recent changes, common causes, and diagnostic value.
- Recommend up to three checks with a purpose and alternative result interpretations.
- Review suspected causes and separate confirmed facts, likely causes, excluded paths, and missing evidence.
- Prepare specialist handoffs when evidence points beyond Kubernetes triage.

## Use Cases

- Prioritize Incident Checks: Turn a production alert, timeline, and selected error lines into a short list of evidence-backed hypotheses and next checks.
- Investigate Release Regressions: Compare symptoms before and after a rollout to identify configuration, image, probe, or resource changes worth checking.
- Review Operational AI Responses: Use the included examples, stress tests, and review templates to assess evidence discipline, diagnostic usefulness, and scope boundaries.

## Prompt Templates

### Beginner: Clarify a Pod Failure

```
Use kubernetes-triage-expert. My Pod is not starting, and I only know its namespace is [namespace]. Ask up to three questions that distinguish the failure stage. Do not assume a root cause or access the cluster.
```

### Intermediate: Review Missing Configuration

```
Use kubernetes-triage-expert. Deployment [name] in [namespace] began failing after rollout [time]. The supplied status is CrashLoopBackOff, and the log says missing REDIS_URL. Rank up to three hypotheses and recommend targeted checks. Request configuration references and key names, not secret values.
```

### Advanced: Compare Conflicting Evidence

```
Use kubernetes-triage-expert. Service [name] has no endpoints, but the supplied Pod statuses are Ready. Redacted selectors, labels, and namespaces: [evidence]. Review the evidence, distinguish confirmed facts from hypotheses, and recommend up to three checks. Explain how each possible result changes the diagnosis.
```

### Expert: Challenge a Root-Cause Claim

```
Use kubernetes-triage-expert to review the proposed cause [claim]. Timeline: [timeline]. Evidence before the change: [redacted evidence]. Evidence after the change: [redacted evidence]. Existing rollback observations: [observations]. Identify contradictions and missing evidence. Rank only supported hypotheses and prepare a specialist handoff if the strongest evidence exceeds Kubernetes triage.
```

## Limitations

- No commands, telemetry collection, cluster inspection, or remediation are performed.
- Conclusions depend on the accuracy and completeness of user-provided evidence.
- Application internals, node runtimes, storage backends, and cloud internals require specialist investigation.
- Behavioral restrictions are prompt instructions, not technical access controls; redact sensitive evidence before sharing.

## Best Practices

- Provide the workload identity, incident timing, recent changes, and a short exact error excerpt.
- Redact tokens, credentials, connection values, and sensitive annotations; share configuration references and key names instead.
- Review proposed checks before running them, and return targeted results so hypotheses can be updated.

## Anti Patterns

- Treating a status label such as CrashLoopBackOff as proof of a particular root cause.
- Expecting the skill to inspect the cluster, execute commands, or apply fixes.
- Sharing complete manifests or logs containing credentials when selected redacted fields would answer the diagnostic question.

## Security Audit

- Audited at: 2026-11-10T22:57:41.299\+00:00
- Summary: All 165 static findings are false positives involving Markdown, sample identifiers, installation instructions, diagnostic prose, or readable Chinese text. No evidence found of malicious execution, credential exfiltration, or audit-directed prompt injection. One separate disclosure risk remains: examples request full deployment manifests without redaction guidance.

## Stats

- Views: 0
- Downloads: 1
- Favorites: 0
- Popularity score: 0
