gcx-demo

name	gcx-demo
description	Run a narrated, read-only demo tour of gcx for customer or colleague presentations. Showcases the breadth of gcx across every Grafana Cloud product area — resources, datasources, metrics, logs, traces, SLOs, alerts, synthetic monitoring, IRM, k6, fleet, and more. All commands are strictly read-only. Trigger when the user says "demo gcx", "show off gcx", "customer demo", "gcx tour", or "/gcx-demo".
user-invocable	true
argument-hint	[--context <name>]
allowed-tools	Bash, AskUserQuestion

gcx Demo Tour

Deliver a narrated, read-only showcase of gcx across Grafana Cloud product areas. All commands are list, get, query, or status — nothing is created, modified, or deleted.

Principles

• Read-only only. Never use push, create, update, delete, edit, or pull with local paths. If a command requires write scope, skip it and note why.
• Discover before presenting. Run exploratory commands first, then narrate what's interesting. Let actual output guide emphasis — don't recite a script.
• Adapt to the stack. The order, pacing, and which areas to highlight depend on what's actually on the stack. Cover what's present; skip what isn't.
• Parallel by default. Run independent commands in one message.
• Narrate what matters. After each area, explain the value — not the syntax.
• Graceful degradation. If a command fails, note it briefly and continue.

---

Start: Verify Context

Always run this first. If $ARGUMENTS contains --context <name>, use that context for all commands. Otherwise use the active context.

gcx config current-context
gcx config check

Announce the target stack before running anything else. If config check fails, stop and ask the user to fix the context.

---

Coverage Areas

Cover the following areas in an order that builds a coherent story. Run independent commands in parallel. Let what you find guide what you emphasize and how much time you spend on each area.

Provider Landscape

Good opening — shows breadth at a glance.

gcx providers list

K8s-native Resources

gcx resources get dashboards -o wide --no-truncate
gcx resources get folders --no-truncate
gcx resources schemas

Dashboards are K8s resources — listable, pushable, validateable. The URL column in -o wide gives a direct deep link for every dashboard. gcx resources schemas reveals the full type catalog including any plugin-installed types (e.g. Adaptive Logs DropRule). gcx resources examples <Kind> produces a ready-to-push template for any of them.

Datasources

gcx datasources list

Every connected datasource — cloud, PDC-tunneled, third-party. All queryable from the CLI.

Live Signals

Discover datasource UIDs first (gcx datasources list -o json), then query whichever signal types are present — skip any that aren't on the stack.

gcx metrics query 'topk(5, count by (job) (up))' -d <PROM_UID>
gcx logs query '{service_name=~".+"}' --limit 5 -d <LOKI_UID>
gcx traces query '{}' --limit 5 -d <TEMPO_UID>

Adapt the queries to what's interesting on this stack — pick label selectors, metric names, or trace filters that will return meaningful output. Mention --open (jump to Grafana Explore) and --share-link (shareable URL) as natural follow-ons.

Assistant

Cloud-only. Requires gcx login (OAuth). Show the commands; run live only if the stack is healthy and the presenter has time — streaming output has variable latency.

gcx assistant prompt "What alerts are firing right now and why?"
gcx assistant prompt "Which service owns checkout-latency?" --continue
gcx assistant prompt "Summarize CPU on prod" --json

gcx assistant investigations list
gcx assistant investigations todos <id>
gcx assistant investigations timeline <id>
gcx assistant investigations report <id>

prompt runs natural language against the stack's live data — the Assistant already has context for dashboards, datasources, and alerts. --continue threads follow-ups via a stored context ID. --json emits a structured event stream for agent tools (Claude Code, Cursor) or scripts.

Investigations are autonomous multi-step LLM runs. The read-only views show plan (todos), chronological activity (timeline), and final findings (report). --open on investigations get deep-links into the Grafana UI.

If investigations list is empty, note it and skip the per-investigation views. Do not create one during the demo.

Reliability & Alerting

gcx slo definitions status
gcx alert instances list --state firing
gcx alert rules list --no-truncate

SLOs return SLI, error budget, and burn rate in one command — scriptable for release gates. Firing alerts include labels, annotations, and runbook URLs. Alert rules expose full PromQL, evaluation timing, and datasource UIDs.

Synthetic Monitoring

gcx synth checks list
gcx synth probes list

HTTP and browser checks from a global probe network — probes list shows regions, coordinates, and capabilities.

IRM

gcx irm oncall schedules list

Who's on-call right now. Pipeable into runbooks and automation.

Cloud Provider Commands

Require cloud.token and cloud.stack. Skip gracefully if not configured, and note what's needed.

gcx k6 load-tests list --no-truncate
gcx fleet pipelines list --no-truncate

k6: load test catalog alongside the Grafana stack — correlate test timing with live metrics. Fleet: full Alloy HCL pipeline configs deployed to collectors via matcher rules.

---

Close

Summarize what was actually shown — areas covered, notable findings (a specific alert, a latency signal in traces, an interesting datasource mix), and the commands used. Adapt the summary to what was interesting on this particular stack. Don't recite a fixed table.

Close with: "One binary, one context, every Grafana Cloud product. All scriptable, all pipelineable — and with --dry-run on mutations for safe GitOps workflows."

---

Error Handling

Situation	Action
config check fails	Stop. Ask user to fix the context before continuing.
Signal datasource not found	Skip that signal type, note it.
cloud.token / cloud.stack missing	Skip k6 and fleet, note what's needed.
Assistant unavailable (self-hosted, OAuth missing, 403)	Skip assistant section, note Cloud + OAuth requirement.
Auth scope missing (403)	Note the missing scope, skip, continue.
Empty list (0 resources)	Report "none found" — not an error; continue.
Any other command error	Print the error summary, skip the section, continue.

Never abort for a single failure. Surface errors at the end.