What problem does it solve? Batch generation of images and audio often fails silently: files land in wrong directories, prompts are lost when context resets, and existing assets get overwritten. This Skill enforces a reproducible pipeline where every asset is specified on disk first, generated through a validated adapter, verified on the filesystem, and listed in a completion report. ## Core Features & Use Cases - Contract-first adapters: Six backend adapters (OpenAI Images, Replicate, fal.ai, ComfyUI, ElevenLabs, Suno unofficial) share one CLI contract with --spec, --out, --dry-run, JSON stdout results, and standardized exit codes. - Reproducible unit specs: Every generation unit is written as a YAML spec before any API call, enabling partial workflows like analyze-only, specs-only, generate-only, and regenerate-N with automatic timestamped backups. - Safety guardrails: Mandatory confirmation gate before spending money, absolute-path enforcement, secret redaction on intake (sk-, r8_, Bearer tokens, cookies), and refusal to overwrite existing non-empty outputs. - Use Case: Turn a short story into a 4-panel comic series: the pipeline analyzes the content, writes four image specs with embedded character descriptions, dry-runs the adapters to check API keys, then generates and verifies each PNG on disk. ## Quick Start Ask the agent to turn your story or brief into a series of image or audio assets using the structured asset pipeline, and confirm the backend and scope at the confirmation gate before generation begins.
