← ClaudeAtlas

visuallisted

Make image, SVG, video or speech files through an OpenRouter model picked from a Jev-ranked, priced list, and store the OpenRouter key once. Use on /1337:visual, "/1337:visual setup", when the user asks to generate, draw, render or narrate something as a file, when a prompt carries a "[1337 visual]" context block from the hook, or when a Jev-backed script reports a missing key.
dimitritholen/1337-claude · ★ 0 · Data & Documents · score 62
Install: claude install-skill dimitritholen/1337-claude
You turn a request for a picture, a vector drawing, a video clip or spoken audio into a file on disk, made by the cheapest OpenRouter model that fits, chosen by the user from a short priced list. You never spend credit without that choice. # The flow 1. **The hook asks.** A UserPromptSubmit hook (`skills/visual/route.py`) watches every prompt. When the prompt hits a word prefilter (image, logo, svg, video, voice, ...) and an OpenRouter key is stored, it asks Jev, TypeSafe's decision model, what the prompt wants (text or code, raster image, vector SVG, video, speech) and, for a visual answer, ranks the six cheapest models of that kind. A prompt can carry more than one modality ("an SVG and a transparent PNG"): every visual modality Jev gives 0.3 or more counts, and each gets its own ranking. A prompt that says transparent, transparency, alpha or "dark and light" ranks only the raster models `catalogue.py` marks `alpha: true` when at least one such model exists, and adds `--transparent` to the raster `generate.py` command. It injects one `[1337 visual]` block. 2. **The user picks.** On that block, before anything else, ask with `AskUserQuestion` exactly as the block says: header "Model" (or one question per modality in a single call, headers "SVG model", "Image model", ...), Jev's pick first and marked Recommended, then cheap to expensive, a price in every label, "Stay with Claude" last. Never twice for one prompt. 3. **You gene