Skills for Developers and SREs

Domain-specific Agent Skills packaged as a Maven artifact, for use with JVM AI agents (Spring AI, LangChain4j) and file-based code assistants (Claude Code, Cursor, Codex) via SkillsJars.

Skills

Skill	Purpose
kubernetes-troubleshooting	Systematic triage of Kubernetes workloads with kubectl — pods, PVCs, networking, nodes, and CSI storage providers. Covers failing workloads and proactive "is the cluster OK?" checks (restart-count and disk-headroom reading). Observe-before-you-act ordering.

Layout

skills/
└── kubernetes-troubleshooting/
    ├── SKILL.md                     # trigger + triage decision layer (<= 500 lines)
    └── references/
        ├── pod-states.md            # phases, container states, exit codes
        ├── kubectl-cheatsheet.md    # inspection commands, read-only vs mutating
        ├── storage-provider-troubleshooting.md  # CSI provider CRDs (Longhorn/Ceph), disk-full, snapshot bloat, backups
        ├── gitops-secrets-and-drift.md          # GitOps sync/drift, operator-synced secrets, credential drift
        ├── cluster-wide-events.md               # node reboot / service restart behind many pods restarting
        └── agent-integration.md                 # driving the skill from an autonomous agent (RBAC, risk tiers)
scripts/
├── link-skills.sh                  # symlink skills into .claude/skills for live local editing
└── check-skill-size.sh             # enforce the SKILL.md line budget (run by `mvn verify`)

Each skill is packaged into META-INF/skills/<org>/<repo>/<skill>/ inside the JAR.

SKILL.md is the triage decision layer and stays lean — its whole body loads into the model's context on every trigger, so depth lives in references/ (loaded on demand). mvn verify fails the build if any SKILL.md exceeds 500 lines; raising the budget is a deliberate edit to scripts/check-skill-size.sh.

Build

mvn verify                     # package + enforce the SKILL.md line budget
jar tf target/*.jar | grep SKILL.md

Use it locally (before publishing)

mvn install

Then depend on it from another project at 1.0.0-SNAPSHOT.

Spring AI / custom JVM agent (classpath, no extraction)

<dependency>
  <groupId>org.springaicommunity</groupId>
  <artifactId>spring-ai-agent-utils</artifactId>
  <version>0.5.0</version>
</dependency>
<dependency>
  <groupId>com.ngaxavi.skills</groupId>
  <artifactId>ngaxavi-skills</artifactId>
  <version>1.0.0-SNAPSHOT</version>
</dependency>

agent.skills.paths=classpath:/META-INF/skills

File-based assistant (extraction)

Add the dependency under the SkillsJars plugin, then:

./mvnw skillsjars:extract -Ddir=.claude/skills

Put that command in your AGENTS.md so the agent runs it before working, and add the extraction directory to .gitignore — it is build output.

Publishing

allowed-tools in each SKILL.md frontmatter must match the corresponding skillsjars.skill.<name>.allowed-tools property in pom.xml; the build fails on mismatch. Once building cleanly, push to a public GitHub repo and submit via the form on skillsjars.com.

A note on trust

A skill is executable instruction delivered through a dependency resolver. Read any third-party skill before trusting it, and treat these the same way — the source is here, in plain Markdown, on purpose.