Self-Service Dev Box via CAIPE: Backstage + ArgoCD + Slack

🛠 Lab from CAIPE 101. Used in Section 5. Course landing page: https://academy.tekanaid.com/courses/caipe-101

🟡 Beta lab. The CAIPE project moves fast (multiple releases per week from main) and this lab pins to CAIPE 0.4.10 + IDP Builder v0.10.2. If you notice drift between the upstream UI and the assignment screenshots, the version pins are the source of truth — file an issue rather than upgrading mid-lab.

CAIPE is most compelling when one user prompt fans out across multiple sub-agents and lands a real change in a real system. That is the shape of the "get me a dev box" self-service narrative: a developer asks for a workspace, the supervisor delegates to Backstage's scaffolder to materialise the template, ArgoCD reconciles the workload onto the cluster, and the Slack sub-agent posts the result back to a channel the developer is watching. Four sub-agents, one prompt, one observable outcome.

In this lab you stand up the full CAIPE stack via CNOE's IDP Builder (kind cluster + Backstage + ArgoCD + Gitea + CAIPE supervisor and sub-agents), inspect the dev-box Backstage software template, send the supervisor the prompt "Create a dev box for me named alpha", trace the request through the Backstage and ArgoCD sub-agents, and verify the Slack sub-agent posts to a local Slack-API-shaped fixture (Flask, port 9006) that mimics `chat.postMessage` so the lab does not need a real Slack workspace or bot token. Inspect the fixture's append-only log to see the exact JSON payload the sub-agent sent.

By the end you will have driven CAIPE's reference self-service narrative through every layer it touches: UI (Backstage scaffolder), GitOps (ArgoCD), supervisor (LangGraph routing), and outbound notification (Slack). The same pattern generalises to other multi-step requests CAIPE can route — "open a Jira ticket and PR for X", "scale service Y and notify the on-call channel", and so on.