THE AGENT SKILLS DIRECTORY
Reusable skills for AI agents. Install any of them into a Cohub world's agent with one command.
npx -y @neta-art/neta-search-cli@latest skills find "<what you need>"
Wrote a skill? Publish it here →
Basic Cohub abilities. Most agents already have these mounted; listed here for reference.
Paste this into your agent's chat — it fetches the skill and follows it.
Use the skill: split-photo-paper-collage. Run `npx -y @neta-art/neta-search-cli@latest skills use @sumino/split-photo-paper-collage --via site` and follow what it prints. Read its complete output, redirecting it to a temporary file first if necessary, and resolve relative paths from the skill directory it prints.npx -y @neta-art/neta-search-cli@latest skills use @sumino/split-photo-paper-collageTurn a photo into a half-photo, half-paper collage where one real element crosses a torn edge and becomes touchable.
# Split Photo / Paper Collage The page is cut in two at exactly the middle. Above: a real photograph of a place, untouched. Below: ivory paper. One element of the photo — a road, a water line, a stair, a cable, a branch — crosses the tear and keeps going, reinterpreted on paper as something you could actually use: the road becomes a piano staircase, the water a sheet you can lift, the branch a swing. A tiny ink figure interacts with it. A few handwritten words in the corner. **Demo:** Hatsune Miku on a dusk backstreet, the road's painted line continuing as piano keys. Art in `assets/demo-cover.webp`. ## What the user gets - One vertical image, two halves at exactly 50%, one continuous scene. - A tear that follows the structure's own shape, not a stock torn-paper edge. - One core structure crossing the seam, transformed into a playable form. - An extremely small ink figure doing something physical to it, and a short handwritten English title. - The Work on the Board. ## When to use - "split photo and paper collage", "the second world inside the photo", "photo above, paper below"; - "上下对半", "撕口", "照片续成纸面", "像杂志内页的拼贴". Do not use for: - two separate pictures or a before/after — the halves must be one scene; - a painted illustration throughout — this needs a photographic top half; - a diptych poster with a title plate — that is a web work. ## Hard rules 0. **Sandbox CLI path.** `node /mods/neta/neta …`. 1. **The photograph is untouchable.** No recolouring, no element swaps, no re-composition; only crop. 2. **Exactly 50/50.** Say so in the prompt, then **measure the result** and fix it with `scripts/fix-seam.sh` if the model drifts (it usually does, 40/60 is common). 3. **One core structure, not three.** Pick the single element that can physically cross the boundary; everything else stays in the photo. 4. **The tear follows the structure.** Where the road meets the edge, the paper tears in the road's width and perspective — never a uniform zigzag ribbon across the page. 5. **The paper half is not a doorway.** No windows, portals, stages, frames or floating mini-illustrations: the structure simply keeps growing downward. 6. **The figure is tiny and physical.** One small ink sketch, hands and feet on the structure, mid-action; never a posed mascot, never a second person. 7. **Six words of title, maximum**, handwritten in the empty paper, matching the interaction — "the street had a chorus", not "Collage Artwork No. 3". 8. **Deliver and place.** An image that never reaches the Board is not shipped. ## Pipeline ### 0. Choose the scene and the structure Look at the request (or the place) and answer one question: **what in this photo can keep going?** Road markings, a stream, a stair, a railing, a cable, a branch, a light beam, a shoreline. Then: **if that thing became touchable, what would it be?** A keyboard, a swing, a slide, a hammock, a ladder of paper steps, a page you turn. If no real photo exists yet, generate one first — a plain photographic plate of the place, no people — or use a generated photo that already reads as photographic. ### 1. Fill the scene file Copy `templates/scene.example.json`: `photo.subject` + 3–4 `photo.details` + `photo.structure`, then `paper.transformation`, `paper.interaction`, `paper.fragments`, `figure`, `seam`, `title`. ```bash node scripts/build-prompt.mjs /tmp/<slug>.json > /tmp/<slug>-prompt.txt ``` It refuses a scene with no structure, no transformation, no figure action, or a title over six words. ### 2. Generate ```bash NETA_DEFAULT_IMAGE_MODEL=gpt-image-2 node /mods/neta/neta generate image \ --param size=1024x1536 --param quality=high \ --work-title "<Title> · Paper Collage" \ --prompt "$(cat /tmp/<slug>-prompt.txt)" ``` `gpt-image-2` handles the two-layer layout, the handwriting and the photo texture best; pass a character reference with `--image` when the figure is a known character. ### 3. Look, then fix the seam by arithmetic — not by re-rolling Open the file and measure the torn edge's y. If it is not half the height: ```bash bash scripts/fix-seam.sh <file> <seam-y> /tmp/final.png ``` This crops only the extra paper, leaving the photograph intact, and prints the resulting ratio. If the handwritten title sat below the new bottom edge, lift it back with a pixel copy (the command is printed at the end of the script) — never retype it. One generative rework is allowed, but a ratio miss is better fixed by cropping than by spending another image. ### 4. Place on the Board ```bash node /mods/neta/.agents/skills/board-ops/scripts/place.mjs \ --title "<Title> · 纸面续景" --work <work.path> ``` ## Files ``` scripts/build-prompt.mjs # scene JSON → prompt, with data checks scripts/fix-seam.sh # crop the paper so the seam lands at exactly 50% templates/scene.example.json # Miku demo data, fully filled assets/demo-cover.webp, demo-art.webp references/art-direction.md, qa.md, publishing.md ``` ## Pitfalls - **40/60 splits.** The model almost always gives the photo less than half. Prompt for it, measure, then crop. - **Title lost in the crop.** Keep the title high enough in the empty paper, or move it back by copying its pixels. - **A stock torn ribbon.** "Irregular torn edge" alone produces the same decorative tear everywhere; you must say the tear follows the structure's width and perspective. - **A portal.** Ask for a doorway once and you get a glowing hole in the paper forever — describe the structure continuing, never an opening. - **A second small illustration.** Fragments are slivers of the *same* photo, pinned right under the tear; nothing else is collaged. - **A big figure.** "Small figure" drifts large; say "extremely small, roughly one-tenth of the paper's height". - **Handwriting garbled.** Six words or fewer; long titles come back misspelled. ## Publishing this skill Maintainer note — see `references/publishing.md`.