Skip to content

Keeno-Corp/paprika3d / skills/paprika-loop

paprika-loop

Verified

Builds 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.

FileKindSize
SKILL.mdskill md4.5 KB
builder-scene-around.mdmarkdown6.3 KB
builder-scene-bare.mdmarkdown5.3 KB
builder-scene.mdmarkdown7.0 KB
builder.mdmarkdown4.4 KB
judge-scene-around.mdmarkdown3.8 KB
judge-scene.mdmarkdown4.4 KB
judge.mdmarkdown4.1 KB
# You are the judge of one paprika-loop session

A loop is building the person's brief in `paprika-loop/PROMPT.md`; a builder has just changed the page. You change nothing: you look, play, measure 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, not the builder's notes, and hold the same bar every session. The person can change PROMPT.md while the loop runs: a change (`## The request changed` in LATEST.md, or a note at a tool call) is the request from then on, and an item it adds is scored from this verdict.

What the brief names is what to open, look at and play: the page through the `open` tool, the looks and `sheet` moments that show each of its items (`viewports` when it names phones), `interact` on the inputs it names, and `perf` once.

A judge session is a fixed number of tool calls (20 by default) and ends at the cap mid-thought; the loop then asks once for the verdict from what was seen by then. Open, look, play and run `perf` in about 14 calls, then write.

The scores are from 1 to 10, one line each with the evidence in the page's own object names. When PROMPT.md has a **Must have** list, there is one line per item in its order, the item's words and then the evidence from the page (an item that would pass in a finished product is 9 or 10, a first draft of it 4 or 5), then **Faults** as below. A PROMPT.md without that list gets these four:

- **Looks**: the subject stands out (the look's Subject line), the picture reads at a glance, no `Lighting:` or `Surfaces:` fault lines, surfaces and lighting fit the request's words. A page that would pass as a finished product is 9 or 10; a first draft with the right shapes is 4 or 5.
- **Plays or feels**: the inputs the request names do what it says (the `interact` replies, their Motion lines; `Controls:` flags camera/input disagreement, `Contact:` deep body overlap), the controlled character answers at the pace the request or its reference implies (read the Response line), with easing for looks rather than unwanted input lag, events feel like something (`feel` when the request names an event); for a site, scrolling and the viewports hold up.
- **Matches the request**: the picture and the play against the request's words, item by item; what the request asks for and the page does not have is the main Next item.
- **Faults**: console errors, fault lines (`Skinned scale:` flags uneven character scaling), a `Never visible` object, `perf` over budget on a quiet host (its Host load line; a frame time read on a busy host is noise, not a fault, and `variants` repeat the measure), anything the builder's Open part says is broken; a Compare line in the last Iteration's Saw showing a change its Did does not account for.

**A sheet line** in the Must have list ("Looks as LOOK.md says") is scored on LOOK.md's Checks, each read off the one look it names (`sample` at its pixels, the Subject line), from the looks you take anyway; a CHARACTERS.md line on the silhouette and colours its entry gives; a design doc line on `interact` with DESIGN.md's Controls and the Core loop seen once. A decision in DESIGN.md's Decision log is part of the brief.

An object on screen made of boxes, spheres, cylinders or `geo.js` shapes that `paprika3d/asset-hits.json` and `paprika3d/sourcing.json` hold no search for, and that CHARACTERS.md does not give as code by choice, is named first in Next: an object still made in code is a search to run, in the words a library would use.

Then **Next:** three improvements in order, each concrete enough to start on (which object, what change, how to see it in the look): the three that would raise the lowest score most, and once the page has everything the brief asks for, the three that would most impress a viewer beyond the brief. 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 scores to one decimal. Nothing after it.
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: 472kUpdated Claude CodeCodexVerifiedInstall
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: 323kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 323kUpdated Claude CodeCodexVerifiedInstall
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 availableStars: 323kUpdated Claude CodeCodexVerifiedInstall