メニュー
Use when connecting to a Platforma instance running in Kubernetes. Covers port-forwarding, endpoint checks, pod/job diagnostics, and a fast “find the issue” triage flow.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | k8s-connection |
| description | Use when connecting to a Platforma instance running in Kubernetes. Covers port-forwarding, endpoint checks, pod/job diagnostics, and a fast “find the issue” triage flow. |
Connect to and debug a Platforma instance running in Kubernetes.
Run these checks in order and stop when the root cause is clear.
# 1) Confirm context/namespace
kubectl config current-context
kubectl get ns
# 2) Deployment and pod health
kubectl get deploy,pods -n <namespace> -o wide
kubectl get pods -n <namespace> --sort-by=.status.startTime
# 3) Recent warnings/errors
kubectl get events -n <namespace> --sort-by='.lastTimestamp' | tail -50
# 4) App logs (current + previous)
kubectl logs deployment/<platforma-deployment> -n <namespace> --tail=200
kubectl logs deployment/<platforma-deployment> -n <namespace> --previous --tail=200
If jobs are involved, immediately add:
kubectl get jobs -n <namespace>
kubectl describe job <job-name> -n <namespace>
kubectl logs job/<job-name> -n <namespace> --tail=200
# Forward debug API and metrics from Platforma deployment
kubectl port-forward deployment/<platforma-deployment> -n <namespace> 9091:9091 9090:9090
After this, use localhost:9091 for debug API and pl-db-cli calls.
# If auth is disabled:
curl -sf http://localhost:9091/debug/health || true
# If service address is used in tests:
kubectl get svc -n <namespace>
If endpoint is unreachable, debug networking/readiness before test logic.
kubectl set env deployment/<platforma-deployment> -n <namespace> \
PL_DEBUG_ENABLED=true PL_DEBUG_PORT=9091 PL_DEBUG_IP=0.0.0.0
kubectl rollout status deployment/<platforma-deployment> -n <namespace>
Re-establish port-forward after rollout.
Requirement: PL_DEBUG_IP=0.0.0.0 is required for port-forward access.
kubectl get deployments --all-namespaces | grep -i platforma
kubectl get pods -n <namespace> -l app=<label>
kubectl describe pod <pod-name> -n <namespace>
ImagePullBackOff / ErrImagePull → wrong image/tag or registry authFailedScheduling / Insufficient cpu|memory → cluster capacity or requests too highCrashLoopBackOff / OOMKilled → app startup failure or memory limits too lowBackoffLimitExceeded on Job → retry exhausted; inspect pod logs/eventsForbidden on kubectl actions (pods/exec, secrets, etc.) → RBAC issuekubectl get appwrappers -n <namespace> 2>/dev/null || true
kubectl get workloads -n <namespace> 2>/dev/null || true
kubectl get pvc -n <namespace>
kubectl describe pod <job-pod> -n <namespace>
Use these when Kueue or PVC/workdir behavior is suspected.
pkill -f "port-forward.*<platforma-deployment>" || true
kubectl port-forward deployment/<platforma-deployment> -n <namespace> 9091:9091 9090:9090
If multiple replicas exist, target a fixed pod to avoid random routing:
kubectl port-forward pod/<pod-name> -n <namespace> 9091:9091
When reporting findings, always include:
pod/job/deploy) and namespaceUse when drawing a diagram of an architecture change for review — a two-panel current-vs-planned (before/after) picture that lets a reviewer see what changed, what's removed, what's new, what held, and the one load-bearing why. Triggers: "arch diagram", "architecture diff", "before/after diagram", "diagram this refactor/migration/change", "visualize the change", "current vs planned", or any request to render an architectural change as a reviewable picture (HTML/SVG).
Use when asked to create a "local gitignore" or ignore files locally without committing the ignore rule. Create a `.gitignore.local` file at repo root and wire it via `git config core.excludesFile`, not `.git/info/exclude`.
This skill should be used when the user asks to "dedup the docs", "prune the corpus", "remove duplicate/redundant paragraphs", "reduce duplication across files", "these files repeat the same information", "consolidate repeated content", "the corpus duplicates topics", "map topics across files", or wants to find and remove duplicated prose across a set of text files (plain prose, markdown, SKILL.md, references, commands, specs).
This skill should be used when the user runs "/prx", "/prx <pr-url>", or "/prx post" — to collect the user's review comments against a diff and post them as an inline GitHub PR review. Trigger on "collect my PR comments", "start collecting comments", "post my comments to the PR", "send the review comments".
Build and sharpen a project's domain model. Use when the user wants to pin down domain terminology or a ubiquitous language, record an architectural decision, or when another skill needs to maintain the domain model.
Manage spec writing, implementation, and bug fixing.