Capstone: Build a Custom CAIPE Sub-Agent and Open a Draft PR

!Lab visual

This is the capstone lab for CAIPE 101. You will build a brand-new sub-agent — `forecast_explainer` — by following CAIPE's contribution model end-to-end at release `0.4.10`:

• Bring up the CAIPE stack on a single Ubuntu host using IDP Builder v0.10.2 (kind + ArgoCD + Gitea + Backstage)
• Fork the upstream layout and copy `ai_platform_engineering/agents/template/` into a new agent directory
• Implement an MCP server with one realistic tool (`explain_forecast`) over a mocked data source
• Register the agent with the supervisor via `AgentCard(...)` in `utils/a2a_common/a2a_server.py`
• Write three eval cases (happy path, ambiguous prompt, out-of-domain) and run them
• Drive the new agent end-to-end through the supervisor on a real prompt
• Open a draft PR to the local Gitea-hosted fork that IDP Builder ships, and generate the would-be Slack post for `#cnoe-sig-agentic-ai`

Why a mocked weather-flavored agent? CAIPE 0.4.10 already ships `agents/weather/` as a reference. Building a *second* weather-flavored agent with a different mocked data source is the safe teaching choice — you learn the contribution mechanics without depending on a third-party API key, and you don't pollute the real upstream repo with a redundant clone. The lab targets local Gitea for the draft PR step (see Task 6) and documents the real-world workflow (PR to `cnoe-io/ai-platform-engineering` on github.com + post in CNCF Slack `#cnoe-sig-agentic-ai`) so you can repeat it for real after the lab.

Key resources:

• CAIPE template sub-agent (0.4.10)
• a2a_common AgentCard registration utility
• CAIPE documentation site
• CNOE SIG Agentic-AI charter