k8s-autopilot

Response formatting templates for K8s Operator coordinator results. Load when you need to present Kubernetes resource lists, pod logs, or state-modifying operation results in polished markdown format.

Manages Alertmanager alert lifecycle and operations via MCP tools. Use when the user asks to triage alerts, create or manage silences, check routing configurations, push test alerts, audit silence changes, or review governance compliance. Also use when the user reports being paged incorrectly, asks about on-call status, or wants to mute noisy alerts — even if they don't mention Alertmanager by name. Do NOT use for Prometheus operations (PromQL queries, metrics, exporters, scrape targets, cardinality, recording/alerting rules) — those belong to the prometheus skill. Triggers on keywords: Alertmanager, alerts, silence, on-call, triage, routing, PagerDuty, Slack notification, mute, suppress, blast radius, receiver, inhibition, maintenance window, expire silence.

Coordinator instructions for PATH A/PATH B planning logic, write_todos usage, and step budgets.

Manages Prometheus monitoring and observability via MCP tools. Use when the user asks to query metrics, onboard applications to Prometheus, install/manage exporters, apply probes, create alerting or recording rules, analyze cardinality, troubleshoot failed targets, or manage scrape configurations. Also use when the user reports metrics are missing, targets are down, or high cardinality — even if they don't mention Prometheus by name. Do NOT use for Alertmanager operations (silences, routing, alert triage, on-call) — those belong to the alertmanager-operations skill. Triggers on keywords: Prometheus, PromQL, metrics, exporter, ServiceMonitor, scrape target, cardinality, alerting rule, recording rule, TSDB, file_sd, remote-write, metric endpoint, up{}, rate(), histogram.

Output formatting and visualization rules for observability subagents. Load this skill when presenting query results (metrics, logs, traces, alerts, or OTel pipeline data) to the user. Teaches when and how to use A2UI interactive dashboards vs Markdown tables, the Dual-Execution pattern for analytical questions, and consistent formatting conventions. Applies to all observability domains: Prometheus, Loki, Tempo, Alertmanager, and OpenTelemetry.

Manages Grafana Loki log observability via MCP tools. Use when the user asks to explore log labels, discover log structure (fields, patterns, parsers), build or execute LogQL queries, estimate query cost, investigate errors from logs, check service health via logs, or correlate logs with traces using trace_id structured metadata. Also use when the user reports missing logs, asks about log formats, or wants to understand the label taxonomy of their Loki cluster. Triggers on keywords: Loki, LogQL, logs, log query, labels, log patterns, log fields, error logs, log structure, trace_id, log rate, service_name, get_cluster_labels. Do NOT use for Prometheus metrics, Alertmanager routing, or OpenTelemetry pipeline operations — those belong to their respective skills.

Manages OpenTelemetry pipelines and instrumentation via MCP tools. Use when the user asks to provision collectors, onboard applications to OTel (auto-instrumentation), validate or investigate pipeline processors, audit metric cardinality (SpanMetrics/Histograms), optimize sampling strategies (head vs. tail), or assess eBPF security posture. Triggers on keywords: OpenTelemetry, OTel, collector, traces, spanmetrics, auto-instrumentation, tail sampling. Do NOT use for Prometheus-only metrics or Alertmanager routing unless directly related to an OTel pipeline.

Manages Grafana Tempo distributed tracing via MCP tools. Use when the user asks to search for traces, investigate errors or latency spikes, analyze trace critical paths, build TraceQL queries, run RED metrics from trace data, explore service topology, correlate traces with logs or metrics, run backend diagnostics, or manage Tempo Operator CRDs (TempoStack / TempoMonolithic). Also use when the user asks about exemplars, spanmetrics, cross-pillar pivots, trace comparison, or generating PromQL alerting expressions from trace patterns. Triggers on keywords: Tempo, traces, tracing, TraceQL, trace_id, trace search, latency, critical path, span, exemplar, spanmetrics, service topology, tempo_traceql_search, tempo_summarize_trace, RED metrics, P99, error triage, missing traces. Do NOT use for Prometheus metrics scraping, Alertmanager silence management, OpenTelemetry collector provisioning, or Loki log queries — those belong to their respective skills.

Manages ArgoCD GitOps workflows via MCP tools. Use when the user asks to deploy, sync, rollback, debug, onboard repositories, manage projects, monitor applications, or perform any ArgoCD operation. Also use when the user reports an app as degraded, out-of-sync, or failing — even if they don't mention ArgoCD by name. Triggers on keywords: ArgoCD, GitOps, sync, rollback, deploy application, Argo application, application health, out-of-sync, app degraded, onboard repo, ArgoCD project, deployment failed, create application, delete.

Use when the user asks to migrate deployments to rollouts, run canary or blue-green deployments, promote or abort a rollout, set up Prometheus analysis templates, monitor rollout health, or perform any Argo Rollouts operation. Also use when the user reports a rollout stuck, canary failing, or deployment not progressing — even if they don't mention Argo Rollouts by name. Triggers on keywords: Argo Rollouts, canary, blue-green, progressive delivery, rollout promote, rollout abort, rollout pause, migrate deployment, workloadRef, AnalysisTemplate, traffic weight, canary step, rollout degraded, rollout phase, experiment, A/B test, rollout stuck, canary failing, deployment not progressing.

Use when the user asks to create or update routes, split traffic, run canary deployments with weighted routing, set up middleware (rate limiting, circuit breakers, auth, IP allowlists), mirror traffic for shadow testing, migrate from NGINX to Traefik, configure TCP routing, manage TLS, enable sticky sessions, or monitor traffic health — even if they don't mention Traefik by name. Triggers on keywords: Traefik, IngressRoute, TraefikService, canary route, traffic split, weighted routing, middleware, rate limit, circuit breaker, NGINX migration, shadow launch, traffic mirroring, TCP route, sticky session, strip prefix, forward auth, ServersTransport, traffic anomaly.

Commits validated Helm chart files to GitHub via MCP server tools. Use when the coordinator delegates a GitHub push after HITL approval and the user has provided a target repository and branch. Also use when updating existing chart files on GitHub. Triggers on keywords: commit, push, GitHub, repository, branch, create_or_update_file, git push, PR, pull request.

Generates complete, production-ready Helm chart files following Bitnami conventions. Use when the coordinator delegates chart generation for any application type. Reads the app-specific skill's SKILL.md and reference files to determine exactly which templates to create, then writes each file to the virtual workspace. Also use when asked to fix helm lint or helm template errors in previously generated charts. Triggers on keywords: Helm chart, generate chart, write templates, values.yaml, deployment.yaml, Chart.yaml, _helpers.tpl, scaffold chart, create chart, Bitnami.

Manages Helm chart lifecycle on Kubernetes clusters via MCP tools. Use when the user asks to install, upgrade, rollback, uninstall, list releases, check status, search charts, or query cluster state related to Helm. Also handles values confirmation, dry-run validation, and installation plan generation. Triggers on keywords: deploy, install chart, upgrade release, rollback, uninstall, helm install, helm upgrade, helm rollback, release status, dry run, list releases, search charts, helm releases, cluster info, chart info, values schema, release history.

Generates per-application skill directories under /skills/ with SKILL.md and reference files that guide the helm-generator sub-agent. Use when the coordinator determines no skill exists for the requested application type AND the helm-planner did not auto-generate skills. This is the fallback skill builder for generic chart generation requests. Triggers on keywords: skill builder, create skill, scaffold skill, new app type, unknown chart.

Fetches existing Helm charts via GitHub MCP, and applies localized surgical edits to templates and values without full rewrites. Use when the coordinator delegates an update to an existing chart that needs patching, version bumping, or targeted configuration changes. Also use when modifying specific template files or values.yaml sections. Triggers on keywords: update chart, patch, modify, bump version, edit values, change template, upgrade chart.

Validates Helm charts by running helm lint and helm template commands in a sandbox. Use when the coordinator delegates chart validation after generation or update. Returns VALID or INVALID with structured error details. Also use when checking chart syntax, template rendering, or values completeness. Triggers on keywords: validate, lint, helm lint, helm template, check chart, verify chart, syntax check.

Use when the user asks to list, get, create, update, or delete any Kubernetes resource; inspect pod logs; exec into pods; run temporary pods; check resource usage (top); manage namespaces; view cluster events; troubleshoot degraded workloads; scale deployments or statefulsets; or manage kubeconfig contexts. Also use when the user reports a pod crashing, a deployment stuck, or a node under pressure — even if they don't mention kubectl by name. Triggers on keywords: kubectl, namespace, pod logs, exec into pod, describe pod, deployment, get pods, apply YAML, delete resource, scale deployment, cluster events, kubeconfig, context switch, top pods, top nodes, CrashLoopBackOff, OOMKilled, pending pod, ImagePullBackOff, cluster health, node stats.