Use when creating a changeset, preparing a release, or bumping versions. Covers which packages to reference, how to write user-facing changeset descriptions, the release automation flow, and the npm/Docker version sync requirement. (project)

Reference for Netlify AI Gateway — the managed proxy that routes calls to OpenAI, Anthropic, and Google Gemini SDKs without provider API keys. Use this skill any time the user wants to add AI on a Netlify site (chat, completion, reasoning, image generation, image-to-image edit/stylize), choose or change a model, wire up the OpenAI / Anthropic / @google/genai SDK, decide which provider to use for an image-gen feature (it's Gemini-only on the gateway), or debug "model not found" / "API key missing" against the gateway. Required reading before pinning a model — the gateway exposes a curated subset, not every provider model.

Use when the user asks to fix open failing reviews, invokes /roborev-fix, or provides job IDs; do not use when the user only pastes review findings with no request to discover or close reviews

Use when the user asks to fix open failing reviews, invokes $roborev-fix, or provides job IDs; do not use when the user only pastes review findings with no request to discover or close reviews

Debug LLM applications using the Phoenix CLI. Fetch traces, analyze errors, structure trace review with open coding and axial coding, inspect datasets, review experiments, query annotation configs, and use the GraphQL API. Use whenever the user is analyzing traces or spans, investigating LLM/agent failures, deciding what to do after instrumenting an app, building failure taxonomies, choosing what evals to write, or asking "what's going wrong", "what kinds of mistakes", or "where do I focus" — even without naming a technique.

Use this skill when answering business analytics, RevOps, GTM metric, pipeline, revenue, funnel, customer, or warehouse questions with Deepline. Triggers on phrases like 'query Snowflake', 'analyze pipeline', 'total ACV', 'break down by quarter', 'use the semantic layer', 'run a semantic query', or any use of snowflake_get_semantic_layer / snowflake_run_semantic_query. Skip prospecting, enrichment, contact finding, outbound, or personalization workflows; use deepline-gtm for those.

Run a dbt Fusion project with Astronomer Cosmos. Use when running a dbt Fusion project with Astronomer Cosmos (Cosmos 1.11+, ExecutionMode.LOCAL on Snowflake/Databricks). Before implementing, verify dbt engine is Fusion (not Core), the warehouse is supported, and local execution is acceptable. Does not cover dbt Core.

Comprehensive DAG failure diagnosis and root-cause analysis with structured investigation and prevention recommendations. Use when deep failure investigation is needed, a DAG fails to import/parse or 'airflow dags list' errors on a file; a task or run is failing and must be diagnosed and fixed; requests like 'why did X fail', 'my dag keeps failing — find and fix it', or fixing a broken DAG so it loads cleanly. For simple 'why did it fail / show logs', the airflow skill handles it directly.

Auto-instrument an application's HTTP / gRPC / DB traffic with Grafana Beyla eBPF — no code changes, no SDK, no restart. Covers requirements (Linux 5.8+ with BTF, CAP_SYS_ADMIN, host PID), language matrix (Go / Java / Python / Ruby / Node / .NET / Rust / C++ / PHP), Docker + Helm + DaemonSet install, port- / process- / Kubernetes-metadata discovery, OTLP traces + Prometheus metrics export, routes decorator (cardinality control), trace sampling, and Grafana Cloud via Alloy. Use when adding observability to a service you can't recompile, instrumenting a closed-source binary, getting RED metrics + spans onto Tempo/Mimir without touching the app, or rolling Beyla as a cluster-wide DaemonSet — even when the user says "zero-code APM", "instrument legacy app", "trace this binary", "eBPF observability", or "no SDK" without naming Beyla.

Build a unified telemetry pipeline with Grafana Alloy — one OpenTelemetry-compatible binary that collects metrics, logs, traces, and profiles and ships to Grafana Cloud / Prometheus / Loki / Tempo / Pyroscope. Covers the Alloy config language (blocks, `sys.env`, component refs), `prometheus.scrape` → `remote_write`, `loki.source.file` + `loki.process` → `loki.write`, `otelcol.receiver.otlp` → `otelcol.exporter.otlp`, `pyroscope.scrape`, K8s / Docker / EC2 discovery, relabeling, modules (`import.file/git/http`), clustering, Fleet Management `remotecfg`, the Alloy UI at `:12345`, and `alloy fmt` / `alloy validate`. Use when writing a `config.alloy`, replacing Grafana Agent / OTel Collector, scraping K8s pods, parsing logs, ingesting OTLP, or debugging "Alloy isn't sending anything" — even when the user says "set up the agent", "write me a scrape config", "drop these logs before sending", or "OTel collector config" without naming Alloy.

Route alerts, run on-call rotations, and drive incidents in Grafana IRM / OnCall — integrations (Alertmanager / Grafana Alerting / generic webhook / PagerDuty), Jinja2 routing + grouping templates, escalation chains (wait → notify schedule → notify team → webhook → auto-resolve), schedules (web + iCal + Terraform `grafana_oncall_schedule`), Slack chatops with Acknowledge/Resolve/Silence, and the P1-P4 incident lifecycle. Use when wiring Alertmanager to OnCall, deciding which team gets paged, building rotations from a Google Calendar / iCal, hooking up Slack notifications, or declaring an incident from an alert — even when the user says "page the platform team on critical alerts", "send Prometheus alerts to Slack", "set up our on-call rota", "escalation policy", or "auto-resolve when the alert clears" without naming OnCall / IRM. Heads up — OnCall OSS is in maintenance mode (archived March 2026); Grafana Cloud users should use IRM.

Set up private network connectivity to Grafana Cloud — AWS PrivateLink, Azure Private Link, GCP Private Service Connect, and Private Data Source Connect (PDC). Provisions VPC endpoints, private endpoints, or PSC forwarding rules per signal type (metrics / logs / traces / profiles); wires Alloy to push to the private DNS endpoint instead of the public one; verifies private DNS resolution + endpoint approval state. Use when sending telemetry to Grafana Cloud without traversing the public internet, eliminating cloud egress costs, meeting PCI-DSS / HIPAA / data-residency compliance, configuring PDC for private data-source queries, or connecting AWS / Azure / GCP workloads to Grafana — even when the user says "stop paying egress", "keep our telemetry off the public internet", or "PCI compliance for Grafana" without naming PrivateLink.

Cut your Grafana Cloud bill by attributing spend to teams and reducing telemetry volume. Covers FOCUS-compliant billing dashboards, cost-attribution labels in Alloy, Adaptive Metrics (cardinality reduction), Adaptive Logs (drop/sample), Adaptive Traces (tail sampling), usage alerts, and an optimization checklist. Use when investigating a high Grafana Cloud bill, attributing observability cost to a team or service, reducing active series / log bytes / trace spans, or setting up usage / quota alerts — even when the user says "our Grafana bill is too high", "who's burning the most metrics", "drop debug logs", "sample our traces", or "alert me before we hit quota" without naming Cost Management.

Continuously profile applications with Grafana Pyroscope and read the result as flame graphs. Covers three instrumentation paths — language SDK push (Go / Java / Python / Ruby / Node / .NET / Rust), Alloy eBPF auto-instrumentation (no code change, requires kernel 5.8+ with BTF), and SDK → Alloy receiver — plus ProfileQL queries, profile types (CPU / memory / allocations / goroutines / mutex), Grafana Cloud Profiles endpoint, and Span Profiles trace-to-profile linking. Use when adding profiling to a service, deploying Alloy as a cluster-wide eBPF profiler, hunting CPU / memory hotspots from a flame graph, comparing two profiles to find a regression, or correlating a slow Tempo trace to its profile — even when the user says "find what's burning CPU", "flame graph this app", "continuous profiling", "heap hotspots", or "why is allocation so high" without naming Pyroscope.

Turn on AI + ML features in Grafana Cloud — Grafana Assistant (NL → PromQL/LogQL/TraceQL, dashboard build, incident investigation, MCP integration), Dynamic Alerting (Prophet forecasting + DBSCAN outlier detection), Sift (8-analysis automated root-cause), Knowledge Graph + RCA Workbench, and the LLM Plugin (OpenAI / Anthropic / Azure / Ollama / vLLM / LiteLLM). Use when you want anomaly alerts without static thresholds, natural-language querying, automated incident investigation, dashboards generated from a sentence, or a managed LLM proxy for plugins — even when the user says "alert when something looks weird", "explain this PromQL", "find the root cause", "make this a dashboard", or "wire Claude into Grafana" without naming any of these products.

Instrument any app with OpenTelemetry and ship metrics / logs / traces to Grafana Cloud or self-hosted Mimir / Loki / Tempo / Pyroscope. Covers SDK auto-instrumentation for Go, Java (Grafana JVM agent), Python (`opentelemetry-instrument`), Node.js, .NET (`Grafana.OpenTelemetry`), Beyla eBPF for zero-code; Grafana Cloud OTLP gateway + Basic-auth (instanceID + API key, base64); env-var config (`OTEL_EXPORTER_OTLP_*`, `OTEL_RESOURCE_ATTRIBUTES`); Alloy / OTel-Collector pipelines; Kubernetes Operator inject-annotations; and head + tail sampling. Use when instrumenting a service, pointing OTLP at Grafana Cloud, switching from Jaeger / Datadog / New Relic, choosing head- vs tail-sampling, or debugging "spans aren't showing in Explore" — even when the user says "auto-instrument my Java app", "send traces to Grafana", "what env vars do I set", "OTLP endpoint", or "Operator inject" without naming OpenTelemetry.

Use when starting any grafana-app-sdk work — scaffolding a Grafana app, initializing a Grafana App Platform app, picking a deployment mode (standalone operator / grafana/apps / frontend-only), wiring app-specific config, or onboarding to the SDK. Covers `grafana-app-sdk` CLI install, `project init` per deployment mode, project layout, the schema-centric workflow (CUE kinds → generated code → reconciler/admission logic), and `SpecificConfig` for env-driven app configuration. Use even if the user just says "make a new app", "Grafana app platform", "kinds and watchers", "operator scaffold", or asks about `apps/` inside the Grafana repo, without naming the SDK explicitly.

Get RED metrics + service maps + frontend RUM + AI/LLM monitoring out of Grafana Cloud — Application Observability (`traces_spanmetrics_*` from OTel traces, p50/p95/p99 latency, exemplar-to-trace, traces-to-logs / profiles), Frontend Observability with the Faro Web SDK (Core Web Vitals, session replay, `pushError`, React + router integration, `TracingInstrumentation` for browser → backend trace correlation), and AI Observability via OpenLIT (token / cost / latency, GPU, hallucination + toxicity evals). Use when standing up APM for a service, wiring an Alloy OTLP receiver + forwarding to Cloud, instrumenting a React frontend for RUM, debugging why service-map edges are missing, monitoring LLM cost drift, or correlating a frontend error to its backend trace — even when the user says "set up APM", "show service map", "monitor browser perf", "session replay", "RUM SDK", or "watch our OpenAI bill" without naming App / Frontend / AI Observability.

Stand up Grafana Tempo as a cost-efficient distributed-tracing backend that only needs object storage, and write TraceQL queries against it. Covers OTLP + Jaeger + Zipkin ingestion, the distributor → live-store → block-builder → object-storage write path, metrics-generator for RED spanmetrics + service graphs, Helm `tempo-distributed` deployment, multi-tenant `X-Scope-OrgID`, TraceQL span / resource / event scopes, structural operators (`>>`, `<<`), `rate()` + `quantile_over_time` metrics, and the traces-to-logs / metrics / profiles datasource links. Use when deploying Tempo, writing a TraceQL query for slow / errored requests, debugging "no traces showing in Explore", sizing queriers / compactors, configuring S3 / GCS / Azure block storage, or wiring trace ↔ log ↔ profile correlation — even when the user says "tracing backend", "find slow requests", "show me the service graph", "store traces in S3", "Jaeger compatible store", or "what called this span" without naming Tempo.

Stand up Grafana Mimir for horizontally scalable, multi-tenant, long-term Prometheus + OTLP metrics storage. Covers monolithic / read-write / microservices deployment, S3 / GCS / Azure / filesystem block storage, Prometheus `remote_write` and OTLP ingestion, multi-tenancy with `X-Scope-OrgID`, ingester replication factor, compactor retention, and per-tenant limits. Use when running Mimir locally or on Kubernetes (Helm `mimir-distributed`), scaling Prometheus past a single node, picking ingest / query / backend split, configuring tenants and ingestion rate, debugging `/ready` 503s or `429 Too Many Requests`, or pointing Grafana at a Mimir datasource — even when the user says "I need long-term Prometheus storage", "scale Prometheus", "multi-tenant metrics backend", "Cortex replacement", "remote_write target", or "store 10M active series" without naming Mimir.