image-deep-researchlisted

# Image deep research Most design questions are answered by looking at twenty examples, not by reading one article about them. This skill gets the pictures in front of you cheaply, checks that they are real, keeps the licence beside every one, and then makes you read them before you say anything about them. The request is whatever the user typed with the skill. If it is empty, ask one question: what should be looked at. ## Where the scripts are The scripts live in `scripts/` next to this file. In Claude Code that folder is `${CLAUDE_SKILL_DIR}/scripts`. In Codex, or anywhere that variable comes through empty, use the `scripts` folder beside this SKILL.md. They need Node 22 or newer and nothing else; rendering needs Chrome, Edge or Chromium (set `IDR_BROWSER` to the executable if it is somewhere unusual). The browser runs headless and never opens a window. ## The loop Deep research here means rounds, not one search: 1. **Write the question down** in a form a picture can answer. "What do luxury watch sites look like" produces mush; "what ground colour, what headline size, and what the first screen shows on six luxury watch sites" produces numbers you can build with. 2. **Gather** from the two sources below: real sites for how a thing is done on the web, open collections for subjects, palettes, materials and art. 3. **Read the contact sheets**, then only the two or three individual images worth a closer look. A sheet of eight costs one read instead of eight