What problem does it solve? Image generation with Codex image_gen often fails silently: prompts say "save as X" but files land elsewhere, stale outputs masquerade as success, parallel sessions grab each other's images, and stdin hangs block execution. This Skill wraps the entire flow—prompt assembly, execution, collection, and completion judgment—into one self-contained entry point so callers never handle these traps themselves. ## Core Features & Use Cases - Fixed 10-field prompt shell: Assembles prompts in a strict labeled order (Use case, Asset type, Style references, Style, House structure, Composition, Content, Color palette, Constraints, Avoid) following the official schema. - Safe-mode execution: Runs codex exec with workspace-write sandbox and network access, never using the forbidden full-bypass flag, and auto-adds --skip-git-repo-check outside git repos. - Deterministic collection: Recovers generated images by session id from generated_images directories, stashes old outputs to .stale, and fails with exit 1 rather than filling shortfalls. - Use Case: A factory skill (e.g., origin-web-design) needs three brand-consistent illustrations; it passes a fields.json with brand style values and three --out paths, and receives verified artifact paths or a clear failure code. ## Quick Start Ask the AI to generate three illustrations by writing a fields.json with the ten prompt fields and running scripts/image_gen.py with --fields and three --out paths.
