visual-qa
Skill120 starsChanged 7 months agoMITView on GitHub
Visual quality assurance — analyze game screenshots for defects, compare against reference, check motion in frame sequences. Supports Gemini Flash (default), native Claude vision, or both with aggregated verdict.
20 other projects use a version of this.
Adopted by
Most recently taken up by
- DevNewbie1826/skillstheir own version, first seen Nov 8, 2026
- claude-world/cf-browsertheir own version, first seen Nov 1, 2026
- toribryan/design-skillstheir own version, first seen Oct 25, 2026
- UXLana/design-skills-packtheir own version, first seen Oct 25, 2026
- Alfred-Lau/skillstheir own version, first seen Oct 20, 2026
- nimadorostkar/Claude-Skills-collectiontheir own version, first seen Oct 13, 2026
Other versions20 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 20
- RandallLiuXin/GodotMaker★ 555same name
- dylanfeltus/skills★ 179same name
- DanielKerridge/claude-code-power-platform-skills★ 61same name
- nimadorostkar/Claude-Skills-collection★ 33same name
- archibate/agent-skills★ 16same name
- claude-world/cf-browser★ 9same name
- UXLana/design-skills-pack★ 8same name
- dev-ik/ai-design-rules★ 3same name
- DevNewbie1826/skills★ 2same name
- wjgoarxiv/litopencode★ 2same name
10 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
We split this skill into the rules it gives an agent
8 rules. 0 are common, 7 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 20 edited versions we read keep these.
7
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 17
Default (Gemini): Run the script below. All queries go to Gemini 3 Flash.
- Distinctive: line 26
No reference, just a question about screenshots → Question mode
- Very specific: line 43
Always pass --log .vqa.log.
- Distinctive: line 47
Read every image file referenced in the arguments using the Read tool.
- Distinctive: line 49
After producing output, append a debug log entry:
- Distinctive: line 60
Run Gemini script, capture output
- Distinctive: line 61
Read all images with Read tool, do native analysis using criteria below
- Distinctive: line 103
Default Godot materials (grey StandardMaterial3D, magenta missing shader)
“Common” is measured: kept by at least 6 in 10 of the 20 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Visual QA
$ARGUMENTS
CRITICAL: Your job is to find problems, not confirm things look fine. Do not rationalize, justify, or explain away what you see. If it looks wrong, report it.
Backend
- Default (Gemini): Run the script below. All queries go to Gemini 3 Flash.
--nativeflag in arguments: Use Claude vision — read every image with the Read tool, analyze directly. Do NOT run the Gemini script.--bothflag in arguments: Run Gemini first, then do native analysis. Aggregate verdicts (details below).
Mode Detection
From the arguments — freeform text with file paths:
- Reference image mentioned + 1 screenshot → Static mode
- Reference image + multiple frames → Dynamic mode — frames are 0.5s apart (2 FPS cadence)
- No reference, just a question about screenshots → Question mode
Gemini Execution
Parse the arguments to construct the command. The script is at ${CLAUDE_SKILL_DIR}/scripts/visual_qa.py.
# Static
python3 ${CLAUDE_SKILL_DIR}/scripts/visual_qa.py --log .vqa.log [--context "Goal: ... Requirements: ... Verify: ..."] reference.png screenshot.png
# Dynamic
python3 ${CLAUDE_SKILL_DIR}/scripts/visual_qa.py --log .vqa.log [--context "..."] reference.png frame1.png frame2.png ...
# Question
python3 ${CLAUDE_SKILL_DIR}/scripts/visual_qa.py --log .vqa.log --question "the question" screenshot.png [frame2.png ...]Always pass --log .vqa.log. Print the script output as your response.
Native Execution
Read every image file referenced in the arguments using the Read tool. Analyze using the criteria and output format below. Never look at code — only images.
After producing output, append a debug log entry:
printf '%s\n' "$(cat <<'LOGEOF'
{"ts":"$(date -u +%Y-%m-%dT%H:%M:%SZ)","mode":"MODE","model":"native","query":"QUERY","files":["FILE1","FILE2"],"output":"FIRST_LINE..."}
LOGEOF
)" >> .vqa.logAggregated Mode (--both)
- Run Gemini script, capture output
- Read all images with Read tool, do native analysis using criteria below
- Produce combined verdict:
- Either says
fail→fail - Either says
warningand neitherfail→warning - Both
pass→pass
- Either says
- Merge issue lists from both, deduplicate by location + description
- Label each issue source:
[gemini],[native], or[both] - Log both outputs to
.vqa.log
Analysis Criteria
Implementation Quality (static + dynamic)
Assets are usually fine — what breaks is how they're placed, scaled, composed:
- Grid/uniform placement when reference shows organic arrangement
- Uniform/default scale when reference shows varied, purposeful sizing
- Flat composition when reference has depth and layering
- Stretched, tiled, or carelessly applied materials
- Objects unrelated to environment (just placed on a flat plane)
- Camera framing doesn't match reference perspective
Visual Bugs
- Z-fighting (flickering overlapping surfaces)
- Texture stretching, tiling seams, missing textures (magenta/checkerboard)
- Geometry clipping (objects visibly intersecting)
- Floating objects that should be grounded
- Shadow artifacts (detached, through walls, missing)
- Lighting leaks through opaque geometry
- Culling errors (missing faces, disappearing objects)
- UI overlap, truncated text, offscreen elements
Logical Inconsistencies
- Impossible orientations (sideways, upside-down, embedded in terrain)
- Scale mismatches (tree smaller than character, door too small)
- Misplaced objects (furniture on ceiling, rocks in sky)
- Broken spatial relationships (bridge not connecting, stairs into wall)
Placeholder Remnants
- Untextured primitives contrasting with surrounding detail
- Default Godot materials (grey StandardMaterial3D, magenta missing shader)
- Debug artifacts (collision shapes, nav mesh, axis gizmos)
Motion & Animation (dynamic mode only)
Compare consecutive frames (0.5s apart):
- Stuck entities (same position/pose across frames when movement expected)
- Jitter/teleportation (large position jumps between frames)
- Sliding (position changes but pose frozen — ice-skating)
- Physics breaks (objects through walls, endless bouncing, unnatural acceleration)
- Animation mismatches (walk anim at running speed, idle while moving)
- Camera issues (sudden jumps, clipping through geometry)
- Collision failures (overlapping objects that should collide)
Output Format
Static / Dynamic
### Verdict: {pass | fail | warning}
### Reference Match
{1-3 sentences: does the game capture the reference's *intent* — placement logic, scaling, composition, camera? Distinguish lazy implementation (fail) from asset/engine limitations (acceptable).}
### Goal Assessment
{1-3 sentences from Task Context. "No task context provided." if none.}
### Issues
{If none: "No issues detected." Otherwise:}
#### Issue {N}: {short title}
- **Type:** style mismatch | visual bug | logical inconsistency | motion anomaly | placeholder
- **Severity:** major | minor | note
- **Frames:** {dynamic only: which frames}
- **Location:** {where in frame}
- **Description:** {1-2 sentences}
### Summary
{One sentence.}Severity: major/minor = must fix. note = cosmetic, can ship.
Question Mode
### Answer
{Direct, specific, actionable answer. Reference locations, frames, colors, objects.}
### Visual Evidence
{What in the screenshots supports the answer. Reference specific frames and locations.}Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

