@snowflake
Agent skills and MCP configuration for deploying and managing Observe's observability infrastructure
Investigate alerts using systematic SRE methodology to understand issues, assess impact, test hypotheses, and identify root causes with evidence-based analysis. Use when: (1) User asks to investigate or analyze an alert (2) User mentions a specific alert or monitor that is firing (3) Debugging production issues or incidents (4) User asks about alert root cause or impact (5) User wants to understand why something is alerting (6) Performing incident response or triage.
Troubleshoot and debug Observe Agent data collection on Kubernetes. Use this skill whenever the user reports any K8s-side problem with Observe — pods crashing, stuck Pending, ImagePullBackOff, "no data in Observe", missing datastreams, suspected token/secret issues, endpoint reachability failures, or platform-specific symptoms on Fargate / GKE Autopilot / EKS Auto Mode — even if they don't explicitly say "debug" or "Observe Agent." Covers pod status, secrets, endpoint reachability, debug logging, self-monitoring metrics, and platform-specific failure modes.
Troubleshoot and debug Observe Agent data collection on Linux hosts. Use this skill whenever the user reports any host-side problem with Observe — the agent failing to start, `observe-agent status` reporting errors, `init-config` failing with unknown flags, "no data in Observe" from a Linux host, fleet heartbeats not arriving, apt/yum repo failures, `unauthorized` token errors, or systemd service failures — even if they don't explicitly say "debug" or "observe-agent." Covers agent status, live log tailing, config validation, debug logging, datastream ingestion verification, and common error patterns.
End-to-end onboarding of a Kubernetes cluster to Observe — creates the backend datastreams + content via the Observe CLI and deploys the observe/agent helm chart. Use for ANY request to install or set up the Observe Agent on Kubernetes, onboard a K8s cluster to Observe, or send K8s data to Observe — across any flavor (EKS, GKE, AKS, GKE Autopilot, EKS Fargate, EKS Auto Mode, minikube, kind, on-prem). Do NOT route a Kubernetes onboarding request to a generic datastream-creation flow; always orchestrate via this skill, which delegates to setup-k8s-backend and setup-k8s-collection.
End-to-end Host Explorer deployment for Observe — creates backend resources (datastreams, content, ingest token) via the Observe CLI and installs the observe-agent on a Linux host. Use this skill whenever the user mentions setting up Observe on a Linux machine, onboarding a host (Ubuntu, Debian, RHEL, CentOS) to Observe, deploying host monitoring from the terminal or CI, or replacing the Observe UI's Host setup wizard — even if they don't say "Host Explorer" explicitly. Orchestrates setup-linux-host-backend and setup-linux-host-collection.
Contains dataset kind selection, column selection rules, core syntax, and the skill index. NEVER generate OPAL, and NEVER invoke an opal-* reference, without first loading this skill.
Use the Observe CLI (`observe`) to investigate production systems and pull telemetry. Reach for this skill whenever the user wants to: search or tail logs; query metrics (CPU, memory, latency, error rate, request volume, custom app metrics); explore traces and spans; correlate events across services, hosts, containers, or Kubernetes resources; investigate or triage alerts (what fired, why, what's still active, related signals); debug an incident or production issue; check service or system health; compare service RED metrics or dependencies; figure out what's happening with a customer, deployment, or user; find which services emit a given tag or label; slice production data by service, environment, region, host, pod, status code, etc.; or pull any observability data out of the platform for analysis, dashboards, or scripting. Also covers the underlying primitives the CLI exposes (datasets, OPAL queries, knowledge-graph tag search, alerts, AI agent skills, auth) when the user asks about them directly.
Installs and configures OpenTelemetry auto-instrumentation for applications written in Java, .NET, Node.js, Python, Ruby. Use when instrumenting applications with opentelemetry auto-instrumentation, auditing existing auto-instrumentation implementations, migrating from vendor locked instrumentations to OpenTelemetry, or checking for latest versions of instrumentation libraries.
Canned queries for various APM features in Observe. Use to validate auto-instrumentation, or when validating or troubleshooting OpenTelemetry instrumentation for a service sending telemetry to Observe.
Identify which field values correlate with bad behavior (slowness, errors, anomalies, unusual values) using phi-coefficient correlation analysis over OPAL. Works on any time-series data — metrics, structured logs, span/trace data, or any dataset where rows can be split into a 'bad' and 'good' cohort by a threshold. Use when: (1) User asks for root-cause analysis on a dataset or metric (2) User wants to know what attributes / dimensions / values are most associated with a failure mode, anomaly, or unusual cohort (3) Investigating which services, hosts, regions, namespaces, or attributes drive outliers (4) User mentions phi coefficient, correlation, or outlier detection (5) User asks 'why is X slow/failing', 'what caused the errors on X', or 'what's different about the bad cohort'.
How to fill in a query card's `visualizationTemplate` (the JSON object that renders a chart). Load this whenever the user asks for a chart, graph, plot, trend, "show me over time", ranking, single number/KPI, scatter, waterfall, or map. Acts as the directory for the per-chart JSON schemas in its references.
Create the Observe backend resources required for Kubernetes Explorer: datastreams, content packages, and an ingest token. Use this skill whenever the user needs to provision Observe backend objects for K8s monitoring — whether they're starting a fresh setup, recreating resources after manual removal, adding a missing datastream, or generating an ingest token for the helm chart — and even when they don't explicitly say "datastream" or "backend." Also invoked by deploy-k8s-explorer during full setup.
Deploy the Observe Agent helm chart on Kubernetes to collect logs, metrics, traces, Prometheus metrics, and K8s events. Use this skill whenever the user mentions installing the observe/agent helm chart, configuring helm values for an Observe agent, deploying observability to a K8s cluster, or any K8s-platform-specific Observe setup (Standard, EKS Fargate, GKE Autopilot, EKS Auto Mode) — even if they don't name the helm chart explicitly. Does not create Observe backend resources (datastreams, tokens) — see setup-k8s-backend for that.
Create the Observe backend resources required for Host Explorer: datastreams, content packages, and an ingest token. Use this skill whenever the user needs to provision Observe backend objects for a Linux host — whether they're starting a fresh setup, recreating resources after manual removal, adding a missing datastream, or generating an ingest token for the observe-agent — and even when they don't explicitly say "datastream" or "backend." Also invoked by deploy-linux-host-explorer during full setup.
Install and configure the Observe Agent on a Linux host. Use this skill whenever the user mentions installing observe-agent, configuring host monitoring on Ubuntu/Debian or RHEL/CentOS, running `init-config` for the agent, upgrading or uninstalling the observe-agent package, or forwarding application OTLP through the host agent — even if they don't name the package or say "host collection" explicitly. Does not create Observe backend resources (datastreams, content, tokens) — see setup-linux-host-backend or deploy-linux-host-explorer for that.