_ ____ _____ ____ ___ ____ _____ ____ _____ ___ ____
/ \ | _ \|_ _| | _ \_ _| _ \| ____/ ___|_ _/ _ \| _ \
/ _ \ | |_) | | | | | | | || |_) | _|| | | || | | | |_) |
/ ___ \| _ < | | | |_| | || _ <| |__| |___ | || |_| | _ <
/_/ \_\_| \_\ |_| |____/___|_| \_\_____\____| |_| \___/|_| \_\
E V I D E N C E > V I B E S
1 0 / 1 0 O R S T O P
Most visual reviews grade intention, effort, or the best screenshot.
This skill grades the current rendered artifact.
It inspects ordinary play, peak pressure, temporal transitions, mobile-scale readability, grayscale hierarchy, physical material integration, UI/world coherence, originality, and production seams. It separates observation from inference, rejects stale screenshots, and forces every score to carry reproducible evidence.
OBSERVE -> CAPTURE -> DOWNSAMPLE -> DESATURATE
| |
v v
DIAGNOSE <- TRACE THE SYSTEM <- TEST THE MOTION
|
v
FIX THE ROOT -> CACHE-BUST -> RECAPTURE -> RESCORE FROM ZERO
|
+--------------------+--------------------+
| |
ALL TEN = 10/10 ANY GAP REMAINS
| |
PASS STOP
There is no “conditional pass.” No “pass with notes.” No score rounded up because the team worked hard.
| # | Area | The real question |
|---|---|---|
| 1 | Composition | Does the eye land on the right event instantly in every state? |
| 2 | Perspective | Do actors, world, camera, shadows, and occlusion inhabit one space? |
| 3 | Shape language | Is the product recognizable without its logo or palette? |
| 4 | Color and light | Does light explain form, depth, emotion, and gameplay? |
| 5 | Characters and enemies | Do silhouettes, acting, mass, and intent survive every pose? |
| 6 | Animation | Are anticipation, impact, recovery, and world response one physical event? |
| 7 | UI | Does the interface belong to the world while remaining effortless to use? |
| 8 | Originality | Is the visual idea ownable—or merely a polished trend template? |
| 9 | Gameplay readability | Can a new viewer identify action, danger, and causality in under one second? |
| 10 | Production quality | Does any seam reveal how the product was assembled? |
Audio/visual alignment is treated as a premium-quality gate whenever sound is in scope.
- 120 px proof — Downsample the complete phone viewport. Player, threat, objective, safe space, dominant event, and primary decision must survive.
- Grayscale proof — Color cannot hide a collapsed value hierarchy.
- Five-beat motion proof — Rest, anticipation, action, impact, recovery. One hero frame cannot pass broken animation.
- Silhouette proof — Internal texture cannot rescue generic anatomy.
- Physical integration proof — Contact, received light, occlusion, world deformation, projection, and material continuity.
- Short-phone proof — Premium UI that collides with the HUD is not premium UI.
- Fresh-build proof — Every verdict resolves to an exact build, state, frame, viewport, and capture.
- AV proof — Onset, transient, tail, hierarchy, headroom, and spatial behavior must support the visible cause.
Missing evidence creates an automatic score cap. Improvement never earns credit from an old screenshot.
The repository uses the portable SKILL.md structure understood by both Codex and Claude Code. There is one source of truth—no duplicated prompt that can drift.
| Agent | Personal skill location | Invocation |
|---|---|---|
| Codex | $CODEX_HOME/skills/<name>/SKILL.md or ~/.codex/skills/<name>/SKILL.md |
$review-award-caliber-art-direction |
| Claude Code | ~/.claude/skills/<name>/SKILL.md |
/review-award-caliber-art-direction |
agents/openai.yaml adds Codex UI metadata. Claude Code safely ignores it and reads the same root SKILL.md plus the same referenced evidence files. Claude Code follows the cross-tool Agent Skills standard; see its official skills documentation.
Clone the repository directly into your Codex skills directory:
git clone https://github.com/jacobtop-tcg/review-award-caliber-art-direction.git \
"${CODEX_HOME:-$HOME/.codex}/skills/review-award-caliber-art-direction"Restart or refresh Codex so the skill catalog is reloaded.
Clone the same repository into Claude Code's personal skills directory:
git clone https://github.com/jacobtop-tcg/review-award-caliber-art-direction.git \
~/.claude/skills/review-award-caliber-art-directionInvoke it directly:
/review-award-caliber-art-direction
Claude Code detects new skills live when its personal skills directory already exists. Restart Claude Code only if that top-level directory was created during the current session.
For project-only use, clone or copy the repository to:
<project>/.claude/skills/review-award-caliber-art-direction/
Use $review-award-caliber-art-direction on this mobile game.
Refuse a pass until all ten areas are honestly 10/10.
In Claude Code, use the same request after the slash command:
/review-award-caliber-art-direction
Review this mobile game. Refuse a pass until all ten areas are honestly 10/10.
Or target a specific failure mode:
Use $review-award-caliber-art-direction to review enemy motion,
physical world integration, 120 px readability, and UI/world coherence.
Use fresh current-build evidence only.
For a release gate:
Run an award-caliber release review.
Inspect all temporal states, grayscale, short-phone fit, and AV sync.
Return PASS only if every required category is proven 10/10; otherwise STOP.
- An exact evidence ledger with build, URL, viewport, state, frame, motion mode, and proof class.
- Ten independently scored categories with automatic caps.
- P0/P1 system defects—not a shopping list of decorative tweaks.
- The smallest high-leverage fix for every score below 10.
- A precise recapture requirement for the next review.
- Seven brutal answers: generic, boring, forgettable, amateur, million-download blocker, award blocker, and 4.8+ rating blocker.
- One unambiguous verdict: PASS or STOP.
review-award-caliber-art-direction/
├── SKILL.md
├── agents/
│ └── openai.yaml
└── references/
├── evidence-protocol.md
├── scorecard.md
└── report-template.md
SKILL.md contains the operating loop. The references hold the deep capture protocol, category thresholds, score caps, and report contract so agents load only what the task requires.
+--------------------------------------------------------------+
| GOOD ENOUGH IS A SHIPPING DECISION. |
| 10/10 IS AN EVIDENCE CLAIM. |
| |
| IF THE PROOF IS STALE, PARTIAL, STATIC, OR CONVENIENT: STOP |
+--------------------------------------------------------------+
Built for mobile games, interactive products, prototypes, motion systems, and any team pursuing work that should feel authored, memorable, premium, and alive.