1
00:00:00,000 --> 00:00:07,000
The story starts with the principal. I ask, whoami, so the session can establish operating identity.

2
00:00:07,000 --> 00:00:15,000
That makes the relationship explicit: I am the principal, and Codex is the agent acting on my behalf.

3
00:00:15,000 --> 00:00:24,000
Next I ask for my preferences. They add behavioral context for memory, Linked Data, and presentation.

4
00:00:24,000 --> 00:00:31,000
With that context established, I invoke the A2A client skill.

5
00:00:31,000 --> 00:00:40,000
The skill discovers what capabilities an endpoint offers, then supports selection of the skill that matches the task.

6
00:00:40,000 --> 00:00:50,000
The endpoint is chosen by the user. The client reads the endpoint's Agent Card instead of relying on hardcoded assumptions.

7
00:00:50,000 --> 00:01:01,000
The discovery table shows the task URL, authentication mode, streaming support, and published skill count.

8
00:01:01,000 --> 00:01:12,000
The skills-by-endpoint table is the crux: it turns an endpoint into a catalog of callable capabilities.

9
00:01:12,000 --> 00:01:22,000
On URIBurner, the selected skill of interest is OpenLink Data Twingler version 2.0.4.

10
00:01:22,000 --> 00:01:34,000
From there, the workflow can continue into authorization and task execution. OAuth and token capture worked in live QA.

11
00:01:34,000 --> 00:02:06,000
The reusable product pattern is simple: establish identity, establish preferences, invoke A2A, discover skills, and select the skill that matters.