GIGA Cloud Learning Skills

This repository packages fourteen public-safe agent skills for Claude Code, Codex, and independent checkout installation. The Chapter 1 to 4 learning system is joined by five practical OpenStack usage skills. Users can understand a task, inspect an existing environment, prepare a change and execute an explicitly approved operation through the same versioned tools.

Skill system

giga-cloud-learning-guide is the cross-chapter learning coordinator. It keeps one case coherent, selects the earliest chapter that owns an unresolved question, enforces handoff contracts, reconciles conflicting findings, and reports architecture, content, syntax, and evidence separately. It is a skill of skills with its own task, not a directory of names.

Scope	Skill	Learner result
Chapters 1 to 4	giga-cloud-learning-guide	Route, mode, handoff continuity, challenge, and final quality envelope
Chapter 1	workload-cloud-fit-assessment	A bounded service and workload profile
Chapter 1	cloud-delivery-model-advisor	A three-axis operating-arrangement comparison
Chapters 1 to 4	cloud-interoperability-explainer	Interoperability, portability, dependency, and exit findings
Chapters 1 to 4	cloud-decision-gap-discovery	Prioritised evidence gaps and stop conditions
Chapter 1	cloud-decision-matrix-steward	A reviewable service-arrangement decision record
Chapter 2	public-service-platform-architecture	A foundation-readiness architecture brief
Chapter 3	iaas-foundation-practitioner	An IaaS baseline plus a portable OpenStack tenant-resource operator kit
Chapter 4	kubernetes-cluster-lifecycle-practitioner	A cluster-runtime baseline plus a portable CAPI/CAPO operator kit

The nine chapter and cross-cutting skills remain the learning system. Five task-focused OpenStack skills add practical tenant operations without turning every API command into a separate skill. Each usage skill can be installed alone with its complete local operator kit.

OpenStack usage toolkit

Need	Skill
Profile/project/region, catalogue, inventory and quota	openstack-environment
VM creation, lifecycle and resize	openstack-compute
Networks, subnets, routers, security rules and floating IPs	openstack-networking
Volumes, attachments, extension and snapshots	openstack-storage
Cross-service fault diagnosis and repair handoff	openstack-troubleshooting

These are tools for existing tenant environments, not only disposable learning labs. A private VM does not require a floating IP or extra volume. Credentials stay in local OpenStack configuration; service availability and permissions are checked per task. Missing permission for one API is reported without inventing or discarding permitted results.

Each skill includes a standalone assets/operator-kit/scripts/openstack_ops.py and local operation contract. It uses OpenStackClient with explicit project and region, read-only inspection, a fixed action catalogue, saved proposals and digest-bound change approval. Inspect and prepare do not mutate cloud resources. Execution checkpoints IDs and reports partial failures rather than automatically replaying or cleaning up. Known controller ownership continues through its owner.

The skill of skills uses a portable operational router for English and German task requests, preserves the case and prior IDs, activates one focused skill and keeps a continuation queue for cross-domain tasks. A routing result is never approval to change state. The Chapter 3 learning baseline and Chapter 4 handoff remain intact.

See usage and installation and executable operation contract. This release checks OpenStackClient 8.2.0 syntax and offline behavior. It does not claim live acceptance on your cloud.

Learning contract

• ask: answer directly and separate transferable concepts, product mappings, assumptions, evidence status, and named sources.
• learn: teach one bounded concept through a concrete service case, then test understanding.
• work: use Spec -> Run -> Validation. Repeat. to produce a chapter handoff without inventing local facts.
• challenge: obtain the learner's claim or attempted answer, then test assumptions, failure paths, ownership, and evidence.
• review: return independent architecture, content, syntax, and evidence verdicts with repair actions.

The machine-readable route and coverage contract is learning-contract.json. Human-readable handoffs are in docs/handoff-contracts.md. Reviewed chapter revisions and known source tensions are in docs/source-and-scope-map.md.

Practical operator paths

The Chapter 3 and Chapter 4 practitioner skills include commit-pinned operator kits under their local assets/operator-kit/ directories. In work mode they support assess, prepare, provision, validate, maintain, and cleanup with explain, prepare, or execute intent.

A practical run always copies the kit to a protected working directory. Offline validation runs before environment values are accepted. Live changes require an approved target, protected credentials, supplied local inputs, a reviewed plan or diff, explicit action approval, bounded validation and owner-aware cleanup. The skill returns an operator_execution record with real checks and evidence status.

Chapter 3 operates resources inside an existing OpenStack project. Chapter 4 establishes and maintains the Kubernetes cluster-lifecycle layer on that IaaS foundation, including the reviewed per-cluster cert-manager and private Contour/Envoy ingress declarations, bounded storage smoke test, scaling, upgrades, Argo CD path and owner-aware cleanup. Neither skill silently expands into OpenStack control-plane installation or the complete Chapter 5 production platform.

Platform integration

• Claude Code: .claude-plugin/plugin.json identifies this repository as a Claude plugin and discovers the folders under skills/.
• Codex: .codex-plugin/plugin.json packages the same skills/ directory. Each skill also has agents/openai.yaml metadata.
• Direct installation: docs/install-and-use.md describes verified checkout installation and the conditions for public-URL installation.

Data and authority boundary

The repository contains distilled, public-safe learning workflows, contracts, and exercises. It does not contain credentials, internal editorial controls, private evidence, organisation-specific topology, or a production approval. A skill may identify local evidence that is required, but it must not invent that evidence or disclose inaccessible provenance.

Validation

python3 scripts/validate_skills.py
python3 scripts/test_learning_paths.py
python3 scripts/test_operator_kits.py
python3 -m unittest discover -s tests -p 'test_*.py' -v

The first command validates package syntax, manifests, frontmatter, references, relationships, and public-safety markers. The second runs deterministic routing and Chapter 1 to 4 coverage tests. The third verifies checksums, portable paths, script and data syntax, and both operator kits through their offline suites. These checks validate the package and its offline behaviour, not a live cloud or cluster.