agentleFS
liangdabiaoliangdabiao/Godogen/.claude/skills/visual-qa/SKILL.md

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

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
  1. Distinctive:

    Default (Gemini): Run the script below. All queries go to Gemini 3 Flash.

    line 17
  2. Distinctive:

    No reference, just a question about screenshots → Question mode

    line 26
  3. Very specific:

    Always pass --log .vqa.log.

    line 43
  4. Distinctive:

    Read every image file referenced in the arguments using the Read tool.

    line 47
  5. Distinctive:

    After producing output, append a debug log entry:

    line 49
  6. Distinctive:

    Run Gemini script, capture output

    line 60
  7. Distinctive:

    Read all images with Read tool, do native analysis using criteria below

    line 61
  8. Distinctive:

    Default Godot materials (grey StandardMaterial3D, magenta missing shader)

    line 103

“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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

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.
  • --native flag in arguments: Use Claude vision — read every image with the Read tool, analyze directly. Do NOT run the Gemini script.
  • --both flag 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.log

Aggregated Mode (--both)

  1. Run Gemini script, capture output
  2. Read all images with Read tool, do native analysis using criteria below
  3. Produce combined verdict:
    • Either says fail → fail
    • Either says warning and neither fail → warning
    • Both pass → pass
  4. Merge issue lists from both, deduplicate by location + description
  5. Label each issue source: [gemini], [native], or [both]
  6. 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.}

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.