notque/vexjoy-agent / skills/meta/grill-jev
grill-jev
VerifiedBroad Jev interrogation: generate up to 50 context-specific questions about a plan, spec, design, code artifact, or any topic — using all primitive shapes and structured forms to surface hidden problems before they ship.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 19.1 KB |
| references/question-battery.md | markdown | 10.3 KB |
| references/visual-plans.md | markdown | 2.6 KB |
references/visual-plans.md
View on GitHub# Grill a visual goal with images
When the plan or artifact has a visual goal ("looks like the reference", "reads clearly on mobile", "feels like a real broadcast"), put pictures in the state. A description of the output lets Jev judge the description, not the output. Shared API, limits, token cost, and routing: [images.md](../../building-with-jev/references/images.md). Image judgments are for steering and review. They are never the only gate.
## Require these in the state
| Item | Form |
|---|---|
| Rendered candidate | `JevImage` of the real output at the size a user sees |
| Reference | `JevImage` of the target, with the owner's rating in text beside it |
| Measured numbers | Text: widths, ratios, timings |
| Owner's words | Quoted verbatim |
If the artifact has no rendered candidate or no reference, that is the first finding. Report it as a gap and stop asking visual questions until both exist.
## Screenshots: `grill-jev.py --image`; custom batteries: `jev_transport`
`scripts/grill-jev.py --image/--images-dir` judges screenshots with a question battery (see SKILL.md, Images and screenshots). For a candidate-versus-reference battery with custom state, write the questions yourself and call `jev_transport.evaluate`. Keep the images in `state` so the request bills them once.
```python
from jev_images import JevImage
import jev_transport
state = {"candidate": JevImage("cand.png"), "reference": JevImage("ref.png"),
"reference_rating": "owner: reference is 9 of 10", "measured": {"arm_px": 18}}
questions = {
"like_reference": {"type": "score", "instructions": "How close is `candidate` to `reference`?", "criteria": LADDER},
"readable": {"type": "noul", "instructions": "Can every label in `candidate` be read at this size?"},
"worse": {"type": "noul", "instructions": "Does `candidate` show anything broken, clipped, or missing that `reference` has?"},
}
answers = jev_transport.evaluate(state, questions, timeout=90)["answers"]
```
- One trait per question, with a separate "anything worse" Noul.
- A Noul in 0.4-0.6 means the question or the picture is wrong. Crop to the area in doubt and ask again.
- Send the candidate and reference sharp: at least 1,080 px on the long side (JPEG of about 512 KB to 2 MB is the sweet spot, per a TypeSafe engineer), and mobile captures at `deviceScaleFactor` 2 or more. Two images at 1280 x 800 cost about 5,400 tokens; accept that rather than shrinking. Stay within the per-request budget in `jev-production-lessons.md`.
- Put structured context beside the pictures (measured numbers, owner's words); a picture alone is the weakest state.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation). | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK). | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |