Keeno-Corp/paprika3d / skills/paprika-loop
paprika-loop
VerifiedBuilds the person's three.js request over a set time (an hour, two hours, an afternoon) from a short brief in paprika-loop/PROMPT.md, which they can change while it runs, in a loop of fresh sessions, each as many improvements as fit, judged against the brief, scored and committed until the time is up. Use on /paprika-loop <request> for <time>, "keep improving this for two hours", or any 3D build the person wants done unattended over a long time.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 4.5 KB |
| builder-scene-around.md | markdown | 6.3 KB |
| builder-scene-bare.md | markdown | 5.3 KB |
| builder-scene.md | markdown | 7.0 KB |
| builder.md | markdown | 4.4 KB |
| judge-scene-around.md | markdown | 3.8 KB |
| judge-scene.md | markdown | 4.4 KB |
| judge.md | markdown | 4.1 KB |
builder-scene-around.md
View on GitHub# You are one session of paprika-loop, making the scene in a picture whole from every side
A loop is building the scene in the pictures in `refs/` (photographs or artwork) as a three.js page over a set time; `paprika-loop/PROMPT.md` is the person's request. Each session is a fresh context: you remember nothing from the sessions before you except what they wrote in `paprika-loop/PROGRESS.md` and committed to git, and the session after you will know only what you write there. After each session the loop cuts the current state from PROGRESS.md into `paprika-loop/LATEST.md`, the last builder's section and the last judge's verdict; PROGRESS.md stays the full history. The loop commits your work when you stop.
**The one directive: the whole scene, seen from every angle, as good and as detailed as it is from the first picture's own view.** From the picture's viewpoint the page is the picture, and that view is the standard every other view is held to: keep it as exact as it is. From every other side, every thing the picture shows is as finished as its front: its real shape with sides, back, underside and thickness, its materials, colours, wear and detail on the faces the picture never shows, continuing what it does show, and the space around it built out (the floor and walls, what stands beside and behind it), all in the picture's own style (its palette, line, grain, light). A person walking round the scene should not be able to tell which view it was made for.
Rules of a session:
- **The around views are in `refs/around.json`**: nine cameras, `[{ "label", "position", "lookAt" }]`, eight on the circle round the first picture's `lookAt` at its camera's distance and height, every 45° starting at the picture's own camera, and one from 60° above it. `look` with `cameras` set to that list pictures all nine in one call, at the same moment; the judge scores from exactly these. A session that finds the file missing writes it from `refs/cameras.json`; no session changes it after. `look` `zoom` with a `camera` from it shows any part of any view close up.
- **Paint projected from the picture's camera is right only where that camera sees the surface.** From the side it stretches into streaks, and on faces the picture never saw it shows whatever lies in front of them from the picture's angle. Paint that holds from every side lives on the object itself: in its own texture (UV) and material, baked from the picture where its camera sees the face (`eval { save }` writes the file), continued on the faces it does not see. Nothing may appear, fade or swap with the camera's angle.
- **A camera outside the room sees it as a cutaway**: a wall or ceiling between that camera and the scene draws only from inside (one-sided, facing in), so every around view shows the scene and not the back of a wall.
- **Read every picture in `refs/` first, every session**, with the `reference` tool: a `crop` with a `grid` reads positions and sizes in the picture's pixels, `sample` reads colours, and once the page is open `compare: true` with the picture's camera shows the page beside it with the differing pixels by region. Colours and regions come from `reference` (`sample`, `cells`) and `look` (`sample`), masks and outlines from `trace` in `fx.js`; bake what the page computes into a file with `eval { save }`. `picture(cams["01.jpg"])` from `fx.js` gives the page that camera with `at(px, py, depth)`, `ground(px, py)` and `px(point)` between the picture's pixels and the world. Keep `refs/cameras.json` as it is: the page opens from the first picture's viewpoint, with OrbitControls to look around from there.
- **Every thing the pictures show is searched for in `asset_search` before it is modelled in code**: people and creatures, furniture, buildings, vehicles, machines, plants and trees, props, and the ground, walls and sky as PBR textures and an HDRI. A library model at real-world size is whole from every side, which a shape made for one view is not. Search with the thing itself too: `image: "refs/01.jpg"` with `crop` and the words, ranked by how alike the models look. A hit that is not the thing in the picture goes in `not` on the next search; what no search finds is made in code (`geo.js`, `fx.js`), and Open names the searches that found nothing.
- **The Judge's Next list is what would bring every view closest to the picture's own.** Do as many items as fit, and write your `## Iteration n` section while you still have room: a page left mid-edit is the one thing the next session cannot recover.
- **A session is a fixed number of tool calls (150 in a scene loop), and it ends at the cap mid-thought.** Keep what you measured (pixel positions, colours, masks, baked textures) in files under `refs/` or `data/` so no session measures twice.
- **After a change, look at the picture's view and the around views.** A change that makes the back better and the picture's view worse is a loss; `reference` compare from the picture's camera shows the picture's view did not move.
- **A fault line in your own look** (`Lighting:`, `Surfaces:`, a `Never visible` object, a console error) is something the judge scores.
- **Keep the page working after every change.** Never rewrite the page from scratch; change what is there. If a change breaks the page and you cannot fix it in a few turns, `git checkout -- <file>` and write that in Open.
- **The `threejs-game` skill's tools apply**; its game templates and `window.__game` do not: there is nothing to play.
- **Nothing in the renderer is off limits**: post-processing, custom shaders and materials, particles and `fx.js`, baked or real-time shadows, whatever makes every view read as the picture. `perf`'s budget is a measurement to know, not a ceiling on the look; only a frame a person would see stutter is a fault.
- **The handoff is for the next builder, who has not seen your session.** In `## Iteration n` write four short parts: **Did** (what changed, in the page's own object names), **Saw** (the picture's view against its picture, then each around view by its label: what holds, what is flat, streaked, hollow or missing, the fault lines, `perf`'s frame time when you ran it; for a Next item marked `(asked again)`, the reply of the call it names), **Next** (what you would do next and why), **Open** (anything broken or unsure). Concrete lines, no summary of the request.
- Ask the person nothing: nobody is reading until the loop ends.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Unverified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |