k8s-animate

Create a Kubernetes explainer video

The output is an MP4 for a lecture: graphical, animated, almost no text, no subtitles.

Argument voiceover=on|off (default on). on narrates with the local Kokoro TTS voice; off makes a silent video and never installs or loads Kokoro.

Argument setup: install only, no video. Do the "Check the environment" step below, then stop.

Two folders. SKILL is the skill's base directory (the folder holding this file). It holds .venv/, icons/, references/ and scripts/. OUT is <the user's working directory>/<topic>, created per video. Scene file, media/ and the final MP4 live in OUT, never in SKILL (updating or removing the skill deletes SKILL). Use absolute paths; do not rely on the current directory.

What ships with this skill

• icons/: the official Kubernetes community icons (labelled set, CC-BY-4.0). Use only these icons for Kubernetes components and resources. Never redraw, recolor into something else, or substitute a generic shape for a component that has an icon here. For concepts with no icon (CSR, certificate, packet, iptables rules), use simple shapes like Doc and Packet in the reference.
• references/node_join.py: a complete, approved example (kubeadm node join). Copy its style, palette and helper classes.
• references/kokoro_service.py: Kokoro-82M plugged into manim-voiceover.
• references/kit.py: shared palette, helpers and the Explainer base scene plus voiced(Scene). New scenes import from it.
• scripts/setup.sh [voiceover=on|off] [--yes]: reports what is missing and prints the install plan (macOS, Arch, Debian/Ubuntu); installs only with --yes.

file	component	file	component	file	component
api	kube-apiserver	pod	Pod	cm	ConfigMap
etcd	etcd	rs	ReplicaSet	secret	Secret
sched	kube-scheduler	deploy	Deployment	pv	PersistentVolume
c-m	kube-controller-manager	sts	StatefulSet	pvc	PersistentVolumeClaim
c-c-m	cloud-controller-manager	ds	DaemonSet	sc	StorageClass
kubelet	kubelet	job	Job	ns	Namespace
k-proxy	kube-proxy	cronjob	CronJob	sa	ServiceAccount
control-plane	control plane node	svc	Service	role / c-role	Role / ClusterRole
node	worker node	ep	Endpoints	rb / crb	RoleBinding / ClusterRoleBinding
hpa	HorizontalPodAutoscaler	ing	Ingress	quota	ResourceQuota
crd	CustomResourceDefinition	netpol	NetworkPolicy	limits	LimitRange

Manim drops the text baked into these SVGs, and api, c-m, sched, kubelet, k-proxy share the same gear glyph. Always label icons with the Icon(name, label) helper.

Workflow

1. Check the environment

Run bash SKILL/scripts/setup.sh voiceover=<on|off>. If it prints Ready., continue. Otherwise it printed what is missing and the plan (first run on this machine): show the plan to the user and ask before installing.

• On confirmation run it again with --yes. It is slow (torch, system packages), so run it in the background or with a long timeout.
• If the plan has a sudo line you cannot type the password: give the user the command to run themselves (! bash SKILL/scripts/setup.sh voiceover=<on|off> --yes).
• Kokoro is free, Apache-2.0 (fine for paid courses), offline and near-commercial quality; if the user does not want it, suggest voiceover=off.

2. Agree the storyboard

Ask for the topic if it wasn't given. Then propose a storyboard in chat and wait for approval, because rendering takes minutes:

• 6–12 beats, each with: a 1–3 word step label, what moves on screen, one spoken sentence.
• Story told through motion: icons in boxes, packets on arcs, color changes for state. Only short labels on screen.
• Facts must be correct for the current Kubernetes version. If unsure about a detail, say so instead of guessing.

3. Write the scene

Model it on references/node_join.py:

• One NARRATION dict (beat → sentence) as a class attribute, one method per beat; Explainer.construct loops them inside with self.voice(line).
• Write OUT/<topic>.py with from references.kit import *, subclass Explainer, and add <Name>Voiced = voiced(<Name>) at the bottom. Beats are def beat(self, i). The silent <Name> and the narrated <Name>Voiced come from the same code.
• Never name the scene file after a stdlib module (operator.py breaks Python's imports).
• Reuse the palette (dark #0b1220, k8s blue #326ce5, gold token, teal network, amber NotReady, green Ready) and the helpers Icon, Packet, Doc, Step, Pill, send.
• Small classes, minimal code.
• Spell terms the way they're spoken in the narration: kubeadm → "kube admin", etcd → "et-cee-dee", kubectl → "kube control".

4. Preview, then render

Run manim as PYTHONPATH=SKILL SKILL/.venv/bin/manim --media_dir OUT/media ... from OUT.

1. Preview: ... -ql <topic>.py <Name>, then extract ~8 frames with ffmpeg into one contact sheet and look at it once. Fix overlaps, clipped labels, and dim or unreadable text.
2. Render final: ... -qh <topic>.py <Name> when voiceover=off, or <Name>Voiced when voiceover=on. Copy OUT/media/videos/<topic>/1080p60/<Name>.mp4 to OUT/<topic>.mp4.
3. With voiceover=on, confirm the MP4 has an audio stream (ffprobe).
4. Open the MP4 when you finish (open on macOS; xdg-open on Linux, or if there is no display just print the path).

Report the absolute MP4 path, the duration, anything to listen for (pronunciation of unusual terms) and remind the user the icons need the credit "Kubernetes community icons, CC BY 4.0" (the outro already shows it). Voice options: af_heart (default), am_michael, am_fenrir; pass one as voiced(<Name>, voice="am_michael").