KubeSense Skills
Observability skills for KubeSense, grouped by telemetry surface.
Skills
| Skill |
Description |
| kubesense-mcp |
The tool layer — connection, auth, the full tool inventory, the WHERE/field-name contract every query skill inherits. Start here. |
| kubesense-logs |
Search and aggregate logs |
| kubesense-traces |
Spans, latency percentiles, error rates, distributed-trace waterfalls |
| kubesense-metrics |
PromQL/MetricsQL over Kubernetes, infrastructure and cloud-provider (AWS/GCP/Azure/Atlas/Confluent/Kong) metrics |
| kubesense-infra |
Cluster inventory: clusters, nodes, pods, workloads, infra failures, recent deploys |
| kubesense-alerts |
List, investigate, and create alert rules; generate import JSON; migrate Datadog monitors |
| kubesense-dashboards |
Create dashboards, or generate preset JSON to import |
Routing
| The user is asking… |
Skill |
| "show me errors / grep the logs" |
kubesense-logs |
| "p99 latency / slow requests / which service is failing" |
kubesense-traces |
| "CPU, memory, disk, saturation, capacity" |
kubesense-metrics |
| "how is my RDS / EC2 / Azure VM / Atlas cluster doing" |
kubesense-metrics |
| "what's running / why is this pod restarting / what changed" |
kubesense-infra |
| "alert me when… / what's firing / convert this Datadog monitor" |
kubesense-alerts |
| "build me a dashboard" |
kubesense-dashboards |
| "this alert fired — what's wrong?" |
kubesense-alerts, then kubesense-infra |
| a tool returned a field-name or WHERE error |
kubesense-mcp |
Prerequisites
Every skill except kubesense-dashboards and kubesense-alerts (which can generate
JSON offline) needs the KubeSense MCP server connected. Those two need it for their
validate and create tools, and for discovering real metric and field names — offline they
fall back to JSON built from names the user supplies. It is served by kubeapi at
/mcp over Streamable HTTP and authenticates with the same credentials as the REST API.
claude mcp add --scope user --transport http kubesense \
https://<your-kubesense-host>/mcp \
--header "x-api-key: <your-api-key>"
See kubesense-mcp for auth alternatives and verification.
Two Rules That Apply Everywhere
Discover before querying. Never invent a field name, metric name, cluster name,
or label. Every surface has a discovery tool; a guessed name either errors or —
worse — returns an empty result that reads like "zero".
Field names are catalog labels, not storage columns. Logs/traces queries take
type, instance, service, method, status_code — not level, pod_name,
app_service, subtype, return_code. The alert engine accepts a different
set. See kubesense-mcp for the contract and
kubesense-alerts for the divergence.
Install
npx skills add kubesense-ai/kubesense-mcp-skills --full-depth -y
[!IMPORTANT]
--full-depth is required. Without it the installer stops at this root SKILL.md
and installs only this index — 1 skill instead of 8 — with no warning.
If you are reading this as the only installed KubeSense skill, that is what happened.
Re-run the command above with --full-depth.
Verify with npx skills add kubesense-ai/kubesense-mcp-skills --list --full-depth, which
should report 8 skills. Or pick individual skills — see the README.
1---2name: kubesense-skills3description: KubeSense observability skills for AI agents — query logs, traces, and metrics from Kubernetes clusters, inspect cluster inventory, and generate alert and dashboard configuration.4---56# KubeSense Skills78Observability skills for KubeSense, grouped by telemetry surface.910## Skills1112| Skill | Description |13|---|---|14| **kubesense-mcp** | The tool layer — connection, auth, the full tool inventory, the WHERE/field-name contract every query skill inherits. Start here. |15| **kubesense-logs** | Search and aggregate logs |16| **kubesense-traces** | Spans, latency percentiles, error rates, distributed-trace waterfalls |17| **kubesense-metrics** | PromQL/MetricsQL over Kubernetes, infrastructure and cloud-provider (AWS/GCP/Azure/Atlas/Confluent/Kong) metrics |18| **kubesense-infra** | Cluster inventory: clusters, nodes, pods, workloads, infra failures, recent deploys |19| **kubesense-alerts** | List, investigate, and create alert rules; generate import JSON; migrate Datadog monitors |20| **kubesense-dashboards** | Create dashboards, or generate preset JSON to import |2122## Routing2324| The user is asking… | Skill |25|---|---|26| "show me errors / grep the logs" | kubesense-logs |27| "p99 latency / slow requests / which service is failing" | kubesense-traces |28| "CPU, memory, disk, saturation, capacity" | kubesense-metrics |29| "how is my RDS / EC2 / Azure VM / Atlas cluster doing" | kubesense-metrics |30| "what's running / why is this pod restarting / what changed" | kubesense-infra |31| "alert me when… / what's firing / convert this Datadog monitor" | kubesense-alerts |32| "build me a dashboard" | kubesense-dashboards |33| "this alert fired — what's wrong?" | kubesense-alerts, then kubesense-infra |34| a tool returned a field-name or WHERE error | kubesense-mcp |3536## Prerequisites3738Every skill except `kubesense-dashboards` and `kubesense-alerts` (which can generate39JSON offline) needs the **KubeSense MCP server** connected. Those two need it for their40validate and create tools, and for discovering real metric and field names — offline they41fall back to JSON built from names the user supplies. It is served by kubeapi at42`/mcp` over Streamable HTTP and authenticates with the same credentials as the REST API.4344```bash45claude mcp add --scope user --transport http kubesense \46 https://<your-kubesense-host>/mcp \47 --header "x-api-key: <your-api-key>"48```4950See [kubesense-mcp](./kubesense-mcp/SKILL.md) for auth alternatives and verification.5152## Two Rules That Apply Everywhere53541. **Discover before querying.** Never invent a field name, metric name, cluster name,55 or label. Every surface has a discovery tool; a guessed name either errors or —56 worse — returns an empty result that reads like "zero".57582. **Field names are catalog labels, not storage columns.** Logs/traces queries take59 `type`, `instance`, `service`, `method`, `status_code` — *not* `level`, `pod_name`,60 `app_service`, `subtype`, `return_code`. The alert engine accepts a **different**61 set. See [kubesense-mcp](./kubesense-mcp/SKILL.md) for the contract and62 [kubesense-alerts](./kubesense-alerts/SKILL.md) for the divergence.6364## Install6566```bash67npx skills add kubesense-ai/kubesense-mcp-skills --full-depth -y68```6970> [!IMPORTANT]71> **`--full-depth` is required.** Without it the installer stops at this root `SKILL.md`72> and installs only this index — 1 skill instead of 8 — with no warning.73>74> If you are reading this as the *only* installed KubeSense skill, that is what happened.75> Re-run the command above with `--full-depth`.7677Verify with `npx skills add kubesense-ai/kubesense-mcp-skills --list --full-depth`, which78should report 8 skills. Or pick individual skills — see the [README](./README.md).
kubesense-ai