SKILL.md output examples use fabricated session data

[pdebruin]

The output format examples added to the skill ("What's new", "Find sessions", "Scaffold", "Session details", "Next steps") use fabricated session codes and metadata — BRK223, BRK224, DEM310, LAB511 with made-up titles and speakers.

Problem

The skill's golden rule is "never fabricate session IDs, speaker names, or schedule data." Having fake examples in the prompt creates a contradiction — the agent could pattern-match on these and surface them as real sessions.

Suggestions

1. Replace fake session codes with obvious placeholders: BRKxxx, [session-code], [Speaker Name]
2. Or use actual sessions from the live catalog (they exist now — 149 sessions)
3. Remove "Create upgrade branch? (y/n)" and "Open in VS Code? (y/n)" — these imply agent capabilities that don't exist today

Context

The examples themselves are good for shaping output quality (few-shot prompting). The issue is specifically the fabricated data, not the concept of having examples.

cc @mikekinsman

[mikekinsman]

Interesting and great catch. Turns out they're not made up, though. As of this message, 3 of the codes and titles are still accurate for Build 2026, but one has had the session title change.

This highlights one aspect of leading up to events: the session names can sometimes change a few times before they're locked and the event starts. So do we keep changing the skills file to keep up with changes to the event plan?

Titles are still in flux for a few more weeks. I don't think we want to rev the skill file every time these titles change, but it also highlights that a session title changing doesn't necessarily mean the session has changed.

In order to be 100% confident that the examples are 100% real, and not have to keep changing the SKILLS file, we'd have to use examples from the past, e.g. ignite or Build2025. I think that's my preference over made up session codes and made up titles.

[mikekinsman]

Pretty sure this was fixed by #8. @pdebruin confirm?