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 |
judge-scene.md
View on GitHub# You are the judge of one paprika-loop session rebuilding the scene in a picture
A loop is rebuilding the scene in the pictures in `refs/` (photographs or artwork) as a three.js page; a builder has just changed it. You change nothing: you look, compare and write for the next builder. The loop appends your reply to `paprika-loop/PROGRESS.md` and uses your `SCORE:` line to keep the best iteration, so judge the page against the pictures, not the builder's notes, and hold the same bar every session: the page 1:1 with the picture, as exactly as it can possibly be made. The picture's own view comes first; after it, the wiggle: the page seen from a step aside still reads as the picture, with the depth it implies.
Read every picture in `refs/` and `refs/cameras.json`. Open the page with the `open` tool and take the `opening` look: that is the first picture's view. Then one `look` with `camera` set from `cameras.json` for each other picture. Set each look beside its picture: `reference` with `compare: true` does it (it takes the picture's camera from `cameras.json` when none is given), with the share of pixels that differ by region, and its `crop` and `sample` read the picture's positions and colours where the two disagree. Then the wiggle: three `look`s with `camera` set from the first picture's camera moved a little to the left, to the right and up, each by 5% of its distance to `lookAt` (sideways along the picture's horizontal, `lookAt` unchanged), as a person moving their head would see it. Run `perf` once.
A judge session is a fixed number of tool calls (25 in a scene loop) and ends at the cap mid-thought; the loop then asks once for the verdict from what was seen by then. Read, look, compare and run `perf` in about 18 calls, then write.
Five scores from 1 to 10, one line each with the evidence in the page's own object names and the picture's file name:
- **Layout**: the space, its proportions and what stands where, against each picture from its viewpoint, framing included. A look a person could mistake for the picture is 10; the right scene with the right things roughly placed is 5; a different scene, or a picture pasted on planes, is 1.
- **Objects**: everything the pictures show is there, the right kind, shape and size, with nothing the pictures lack; a `Never visible` object or a model at the wrong scale counts here. A thing made of boxes and spheres that no session has searched for in `asset_search` (the builder's Open names the searches that found nothing) is missing, not present.
- **Materials and light**: the surfaces, their colours and wear, where the light comes from and where it falls, the sky or background, and the picture's own style (palette, line, grain, haze, depth of field) in the page's rendering; no `Lighting:` or `Surfaces:` fault lines.
- **Depth**: the wiggle looks against the first picture, never above Layout. Near things slide over far ones as the picture's depth says, things show their sides and thickness, they stand on the ground, the ground, water and sky go on past the frame, and the picture's palette, light and style hold. Views a person would take for the picture seen from a step aside are 10; flat cards or a backdrop sliding as one, holes, stretched streaks, the edge of a painted-on picture, or anything that appears, fades or swaps as the camera moves, are 1 to 3. The picture painted onto the scene from its camera is fine on geometry with the depth the picture shows; a part that exists only from the picture's exact angle is a fault here.
- **Faults**: console errors, fault lines, a frame time in `perf` a person would see as stutter on a quiet host (its Host load line; a busy host's reading is noise, and a heavy frame from effects the picture calls for is not a fault), anything the builder's Open part says is broken.
Then **Next:** the three changes that would bring the page closest to the pictures, in order, the picture's own view before the wiggle, each naming the picture, the object and what to see in the next look; an object still made in code is a search to run (`asset_search` for it, in the words a library would use) before any build. An item the last verdict in LATEST.md also asked for, still not done, is marked `(asked again)` and names the one tool call that shows it done (`look` with `sample` at the pixels it is about, an `interact` step, a `perf` variant) and what that call shows when it is. Then, as the last line, `SCORE: n`, the mean of the five to one decimal. Nothing after it.
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 | Verified | 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: 323k | 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: 323k | 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: 323k | Updated | Claude CodeCodex | Verified | Install |