DEMO2_STORYBOARD

Demo 2 Storyboard — Conversation → Foundry → gxwf → Workflow

3-minute (target) edited YouTube embed for Block C of the GCC2026 talk. Same application (MRSA mobile-AMR) as demo 1. Mechanism: Foundry interview→Galaxy pipeline + gxwf validation. No Galaxy server in the loop until the very end (and optionally not even then).

Status: storyboard. Foundry interview→Galaxy pipeline is in development. Must check status before recording.

Hard constraint

Do not mention Nextflow conversion. The Foundry mode demonstrated here is the interview-driven pipeline. Other Foundry pipelines (nextflow-to-galaxy, cwl-to-galaxy, paper-to-galaxy) exist but are off-topic for this video.

Source materials to reuse

Scenes (target durations)

#LengthSceneNotes
10:15Title card: “Same problem, no Galaxy server. Yet.”Set the contrast with demo 1 immediately.
20:30A terminal or chat surface; user invokes the Foundry interview pipeline. Agent asks targeted questions: “What organism? What inputs? Comparing isolates?”This is the “agent skill compiled from curated knowledge” beat.
30:30Agent emits an intent document — small structured summary (organism, isolates, ARG questions, mobile-context dimensions). Schema-validated; provenance to source Mold visible.The compile-time-grounded artifact.
40:30Cast skills decompose the intent into Format2 step proposals: staramr, Bakta, ISEScan, IntegronFinder, table conversions.Show the Mold-by-Mold authoring.
50:30gxwf validates each emitted step against the ToolShed-served schema for the named tool. One deliberate “bad” parameter triggers a diagnostic with the legal options listed. Agent corrects.The deterministic inner-loop feedback. The “milliseconds vs. minutes” point.
60:20Final Format2 file opens in VS Code (galaxy-workflows-vscode extension). Completions, hover docs, no diagnostics.Visible parity with demo 1’s workflow.
70:15Optional: convert to native .ga and run against a Galaxy server with the MRSA inputs — same workflow, executes.Skip if pacing is tight; the talk’s argument doesn’t require execution.
80:10Cut to side-by-side: workflow from demo 1, workflow from demo 2. Same graph.The synthesis moment that slide D1 will echo.

Voiceover script (rough)

Opening (Scene 1): “Same MRSA question. This time, no Galaxy server. We’re going to build the workflow before we run it.”

Foundry interview (Scenes 2–3): “An agent walks me through the analysis. It’s not free-text RAG — it’s a compiled skill from a curated knowledge base. The intent document it produces has explicit schemas and provenance back to its source materials.”

Mold authoring + gxwf (Scenes 4–5): “Each step is authored by a Mold and validated by gxwf — not at runtime, in milliseconds. Names, types, select options, conditional branches, collection semantics. When the agent gets something wrong, the feedback is structured and specific.”

IDE handoff (Scene 6): “The Format2 file looks the same in VS Code as it does coming out of Foundry. Completions, hovers, diagnostics — same validator, no Galaxy server required.”

Synthesis (Scene 8): “Same workflow as demo one. Two paths in. One reproducibility substrate.”

Editing notes

Open production decisions