Kstack: Skill Pack for Claude Code to Monitor and Troubleshoot Kubernetes
Kstack is a skill pack for Claude Code (and other AI coding agents) that exposes a set of slash commands for monitoring and troubleshooting Kubernetes clusters. It hooks into your existing kubeconfig and RBAC permissions, runs standard tools like kubectl, Kubetail, Trivy, Pluto, and hands the results to the LLM for analysis — aiming for fast, token-efficient responses.
Key Skills
/cluster-status— health snapshot (pod restarts, node conditions, resource pressure)/events— recent events ranked by severity/investigate— root-cause analysis across events, logs, and related resources/logs— shared tmux session translating natural language into log fetches (via Kubetail)/metrics— CPU, memory, and other resource metrics for pods, nodes, workloads/exec— shared tmux shell into a pod, node, or ephemeral debug container/audit-security— RBAC, pod security posture, privilege tightening/audit-network— NetworkPolicy, Service, Ingress, GatewayAPI, DNS, and encryption checks/audit-cost— requests vs. usage, over-provisioning, idle capacity/audit-outdated— outdated services, known CVEs, available version bumps/cleanup— removes all kstack-owned resources (debug containers, pod clones, watcher jobs)/forget— clears local cache and cluster knowledge
Installation
Global install:
curl -sS https://kstack.sh/install | bash
Local project install:
curl -sS https://kstack.sh/install | bash -s -- --local
By default, skills are prefixed with kstack-*; use --no-prefix to disable. The installer auto-detects agents on your PATH and installs for each. Supported agents include Claude Code, OpenAI Codex CLI (--agent codex), OpenCode (--agent opencode), Cursor (--agent cursor), Factory Droid (--agent factory), Slate (--agent slate), Kiro (--agent kiro), and Hermes (--agent hermes).
How It Works
All skills are read-only by default — mutations require explicit confirmation. They honor your local kubeconfig context and RBAC. If permissions are insufficient, kstack reports the issue. Global flags like --context <ctx> are supported across all skills.
Who It's For
Developers and platform engineers who use AI coding agents and need quick, natural-language-driven monitoring and debugging of Kubernetes clusters without switching contexts.
📖 Read the full source: HN LLM Tools
👀 See Also
Indie dev deploys full game studio site via Claude Code, including Steam API data layer
An indie game developer used Claude Code to build and deploy a game studio website without touching a terminal, including a data layer that pulls live info from the Steam API.
Claude AI Session Compaction Issues and Workarounds
Default compaction in Claude AI sessions can degrade retrieval accuracy from ~9.75/10 to ~5/10, causing hallucinations. The user tested with 418K tokens and found manual compaction using Opus maintains accuracy while default compaction fails.
Recall: A Persistent Memory MCP Server for Claude Code
Recall is an open-source MCP server that gives Claude Code persistent memory across sessions via semantic search with embeddings. It includes four lifecycle hooks: session-start, observe, pre-compact, and session-end.
Signet: Open-Source Memory Layer for AI Coding Agents Hits 80% F1 on LoCoMo
Signet is an open-source memory system for AI coding agents that achieves 80% F1 on the LoCoMo benchmark, compared to 41% for standard RAG. It extracts memories after each session and injects relevant context before prompts, running locally with SQLite.