Scenario interfaces and feature coverage¶
Each numbered walkthrough has its CLI commands, an agent prompt, a concrete MCP starter call and console directions. Use the same selected cloud/environment, inputs, expected results and cleanup steps in every interface. Reports and side effects come from the same implementation; choosing an agent does not make a cloud deployment credential-free.
The generated command matrix maps every CLI command/subcommand to a walkthrough. The checked interface examples and coverage manifest bind those walkthroughs to real MCP schemas and catalog items. They describe documented coverage, not a claim that all charts or all cloud deployments have been run live.
One-time setup¶
The human installs Cloudseed and any required runtime, logs in to the provider, authorizes OS keychain access where applicable, and starts the interface. These are authentication or host-control steps rather than actions an already connected MCP server can use to create its own connection.
Select your agent/model and connect the MCP client as shown in 14. Open the console as shown in 15. Enter secrets into the provider login or Credentials form, not chat or operation parameters. Interactive SSH, k9s, browser authentication and desktop VPN prompts may still require a human; noninteractive checks and infrastructure steps have tool/form routes.
Command-to-interface map¶
For MCP, read the named tool's schema (cs mcp tools). A CLI subcommand usually becomes an action field; scan uses kind, managed platforms use service plus args, and every new operation has its own generated tool. Pass cloud and environment explicitly for changes. confirm:true approves the named change, not future unrelated changes.
| CLI family | MCP tool | Console route |
|---|---|---|
setup, plan, apply, destroy | cloudseed_setup, cloudseed_plan, cloudseed_apply, cloudseed_destroy | Create; selected Environment actions |
list, status, output, inventory, troubleshoot, doctor | matching cloudseed_<command> | Overview, Environments, All actions |
update-ip, provision, noninteractive ssh | cloudseed_update_ip, cloudseed_provision, cloudseed_ssh | Selected Environment / All actions |
k8s, env, node | cloudseed_k8s, cloudseed_env, cloudseed_node | Environment Kubernetes/Nodes; current environment selector |
kubectl, helm | cloudseed_kubectl, cloudseed_helm | All actions; explicit argument list |
platform | cloudseed_platform | Platform catalog, plan/install/uninstall/status/UI actions |
vpn | cloudseed_vpn | Environment VPN actions; host prompts may still be required |
finops | cloudseed_finops | All actions → FinOps; Reports |
dr, chaos, scan | cloudseed_dr, cloudseed_chaos, cloudseed_scan | Resilience; Reports |
evidence list\|read | cloudseed_evidence | Reports; full JSON/Markdown; agent task box for paginated review |
usage report | cloudseed_usage | Agents & MCP usage summary; JSON report |
usage install | cloudseed_usage_install, with confirm:true after approval | Confirmed optional analyzer installation action |
databricks, snowflake | cloudseed_managed | All actions → managed platform forms |
undo, explain, help, skill show | cloudseed_undo, cloudseed_explain, cloudseed_help, cloudseed_skill | All actions / Help / context explain buttons |
ops ACTION | cloudseed_ops_ACTION (hyphens become underscores) | All actions → Operations & readiness; Resilience shortcuts |
| Agent/model/service setup, credentials entry, installations | Human host setup; read-only status remains available | Agents & MCP, Credentials, Help; native authentication/installation when requested |
Interactive ssh / k9s | Use noninteractive SSH or kubectl checks; interactive session remains human | Use terminal for the interactive application |
An agent with the bundled skills uses these same operations or their CLI equivalents. Ask for the scenario number, selected environment and exact scope. If the agent cannot authenticate or a prerequisite is missing, complete that step and resume; do not ask it to read credential files or weaken the guard to get a passing result.
Catalog coverage¶
The catalog contains optional and provider-specific items. A group walkthrough teaches the inspect → plan → install → verify → uninstall loop; it does not ask you to install every heavyweight item simultaneously. For each item in the manifest, substitute its name in that loop, review only, requirements and notes via platform info, and provide the resources/credentials it needs. The dependency plan is authoritative.
cs platform info minio
cs platform plan minio vmware --env lab
cs platform install minio vmware --env lab
cs platform status vmware --env lab
cs platform uninstall minio vmware --env lab
MCP performs the same loop with cloudseed_platform, items:["minio"], explicit cloud/env, and action:"info"|"plan"|"install"|"status"|"uninstall"; mutation calls need confirm:true. In the UI select the same item in Platform, inspect its details, Plan, then Install or Uninstall and check Activity. An agent can follow this loop for a specific item or group; it should retain the item-specific notes and verification checks.
| Catalog group | Walkthrough | Additional validation |
|---|---|---|
basek8s, scaling | 06, 13 | Private service access, metrics and node/autoscaler behavior |
data, ai, agentic | 07 | Storage, database/model readiness and application-specific checks |
finops | 15 | OpenCost data requires a running cluster and metrics |
devsecops, security | 06, 10 | Token prerequisites, policy audit results and affected-workload review |
resilience | 08, 18 | Completed backups/restores, volume evidence and recovery objectives |
chaos | 09 | Explicit experiment target, recovery and cleanup |
What is tested¶
Documentation tests parse the CLI examples, validate each checked MCP starter against its schema, require agent/UI instructions on every page, and compare the feature manifest with the current command registry, operations and platform catalog. The executable scenario scripts test local/fixture behavior and clearly skip unavailable live steps. Provider mock tests validate real Terraform schemas without cloud calls. Container and standalone-binary pipelines test their own runtime packaging.
Cloud acceptance is a separate opt-in activity in 19. Without sandbox credentials, a report must say that provisioning, egress, chart readiness, upgrade and recovery were not verified in a live cloud.