# K-01 — Modell vs. Harness
# Category: KEY
# Kapitel: 01 – Einführung
# Anker: 1.2 Die wichtigste Unterscheidung: Modell vs. Harness
# Orientierung: 16:9 landscape
# Didaktik: Next-Token-Modell schlägt vor; Harness führt Tools aus, prüft Permissions/Modes, speist Ergebnisse zurück.
# Caption (DE): Das Modell schlägt vor; der Harness prüft, führt aus und speist Ergebnisse zurück.
# Language: English (Gemini / Imagen)
# Usage: paste entire prompt body below into Gemini / image_gen

--- PROMPT START ---
Educational technical book figure, 16:9 landscape, flat vector with soft 3D depth, deep night navy background, cyan and soft violet accents, warm off-white panels, high contrast, generous margins.

Left third: rounded module labeled exact text "MODEL" with a soft geometric neural-brain icon and violet glow; small chips leaving it labeled exact text "propose".
Center: a single traffic-light permission node labeled exact text "gate" (red / amber / green dots only).
Right third: stacked control-room panel labeled exact text "HARNESS" with three rows labeled exact text "Tools", "Permissions", "Modes"; under Modes three chips exact text "Normal", "Plan", "Always-approve".
Closed loop arrows: propose → gate → execute on the harness → exact text "result" arrow returning into the model.

Exact labels only, spell exactly: "MODEL", "HARNESS", "propose", "gate", "execute", "result", "Tools", "Permissions", "Modes", "Normal", "Plan", "Always-approve".
No logos, no watermarks, no terminal gibberish, no fake code, no meta notes, no palette names on the image. Calm professional practice-book style.
--- PROMPT END ---
