@haposoft/cafekit 0.13.4 → 0.14.0

package/src/claude/skills/ai-multimodal/references/vision-understanding.md

@@ -25,9 +25,9 @@ Comprehensive guide for image analysis, object detection, and visual understandi
 ## Model Selection
 
 ### Gemini 2.5 Series
-- **gemini-2.5-pro**: Best quality, segmentation + detection
-- **gemini-2.5-flash**: Fast, efficient, all features
-- **gemini-2.5-flash-lite**: Lightweight, all features
+- **gemma-4-31b-it**: Best quality, segmentation + detection
+- **gemma-4-31b-it**: Fast, efficient, all features
+- **gemma-4-31b-it-lite**: Lightweight, all features
 
 ### Feature Requirements
 - **Segmentation**: Requires 2.5+ models
@@ -49,7 +49,7 @@ with open('image.jpg', 'rb') as f:
     img_bytes = f.read()
 
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=[
         'Describe this image in detail',
         genai.types.Part.from_bytes(data=img_bytes, mime_type='image/jpeg')
@@ -62,7 +62,7 @@ print(response.text)
 
 ```python
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=[
         'Classify this image. Provide category and confidence level.',
         img_part
@@ -74,7 +74,7 @@ response = client.models.generate_content(
 
 ```python
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=[
         'How many people are in this image and what are they doing?',
         img_part
@@ -88,7 +88,7 @@ response = client.models.generate_content(
 
 ```python
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=[
         'Detect all objects in this image and provide bounding boxes',
         img_part
@@ -103,7 +103,7 @@ response = client.models.generate_content(
 
 ```python
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=[
         'Create a segmentation mask for all people in this image',
         img_part
@@ -122,7 +122,7 @@ img1 = PIL.Image.open('photo1.jpg')
 img2 = PIL.Image.open('photo2.jpg')
 
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=[
         'Compare these two images. What are the differences?',
         img1,
@@ -135,7 +135,7 @@ response = client.models.generate_content(
 
 ```python
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=[
         'Extract all visible text from this image',
         img_part
@@ -155,7 +155,7 @@ with open('image.jpg', 'rb') as f:
     img_bytes = f.read()
 
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=[
         'Analyze this image',
         types.Part.from_bytes(data=img_bytes, mime_type='image/jpeg')
@@ -171,7 +171,7 @@ import PIL.Image
 img = PIL.Image.open('photo.jpg')
 
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=['What is in this image?', img]
 )
 ```
@@ -184,12 +184,12 @@ myfile = client.files.upload(file='large-image.jpg')
 
 # Use multiple times
 response1 = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=['Describe this image', myfile]
 )
 
 response2 = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=['What colors dominate this image?', myfile]
 )
 ```
@@ -198,7 +198,7 @@ response2 = client.models.generate_content(
 
 ```python
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=[
         'Analyze this image',
         types.Part.from_uri(
@@ -250,7 +250,7 @@ class ImageAnalysis(BaseModel):
     scene_type: str
 
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=['Analyze this image', img_part],
     config=genai.types.GenerateContentConfig(
         response_mime_type='application/json',
@@ -272,7 +272,7 @@ images = [
 ]
 
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=['Analyze these images and find common themes'] + images
 )
 ```
@@ -284,7 +284,7 @@ before = PIL.Image.open('before.jpg')
 after = PIL.Image.open('after.jpg')
 
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=[
         'Compare before and after. List all visible changes.',
         before,
@@ -300,7 +300,7 @@ reference = PIL.Image.open('target.jpg')
 candidates = [PIL.Image.open(f'option{i}.jpg') for i in range(5)]
 
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=[
         'Find which candidate images contain objects similar to the reference',
         reference
@@ -333,7 +333,7 @@ response = client.models.generate_content(
 **Few-shot examples**:
 ```python
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=[
         'Example: For an image of a cat on a sofa, respond: "Object: cat, Location: sofa"',
         'Now analyze this image:',
@@ -371,7 +371,7 @@ response = client.models.generate_content(
 
 ```python
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=[
         '''Analyze this product image:
         1. Identify the product
@@ -388,7 +388,7 @@ response = client.models.generate_content(
 
 ```python
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=[
         'Extract all text and UI elements from this screenshot',
         img_part
@@ -400,7 +400,7 @@ response = client.models.generate_content(
 
 ```python
 response = client.models.generate_content(
-    model='gemini-2.5-pro',
+    model='gemma-4-31b-it',
     contents=[
         'Describe visible features in this medical image. Note: This is for informational purposes only.',
         img_part
@@ -412,7 +412,7 @@ response = client.models.generate_content(
 
 ```python
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=[
         'Extract data from this chart and format as JSON',
         img_part
@@ -424,7 +424,7 @@ response = client.models.generate_content(
 
 ```python
 response = client.models.generate_content(
-    model='gemini-2.5-flash',
+    model='gemma-4-31b-it',
     contents=[
         '''Analyze this scene:
         1. Location type
@@ -451,7 +451,7 @@ def analyze_image_with_retry(image_path, prompt, max_retries=3):
                 img_bytes = f.read()
 
             response = client.models.generate_content(
-                model='gemini-2.5-flash',
+                model='gemma-4-31b-it',
                 contents=[
                     prompt,
                     genai.types.Part.from_bytes(

package/src/claude/skills/ai-multimodal/scripts/check_setup.py

@@ -302,7 +302,7 @@ def main():
         print("\nImage generation models:")
         print("  • gemma-4-31b-it-image    - Nano Banana Flash (DEFAULT - fast)")
         print("  • imagen-4.0-generate-001   - Imagen 4 (alternative - production)")
-        print("  • gemini-3-pro-image-preview - Nano Banana Pro (4K text, reasoning)")
+        print("  • gemma-4-31b-it-image-preview - Nano Banana Pro (4K text, reasoning)")
         print("\nExample (uses default model):")
         print("  python scripts/gemini_batch_process.py --task generate \\")
         print("    --prompt 'A sunset over mountains' --aspect-ratio 16:9 --size 2K")

package/src/claude/skills/ai-multimodal/scripts/gemini_batch_process.py

@@ -8,8 +8,8 @@ Supports all Gemini modalities:
 - Video: Summarization, Q&A, scene detection
 - Document: PDF extraction, structured output
 - Generation: Image creation via Imagen 4 or Nano Banana (Gemini native)
-  - Nano Banana 2 (gemini-3.1-flash-image-preview): Fastest, 95% Pro quality (default)
-  - Nano Banana Pro (gemini-3-pro-image-preview): Quality/4K text/reasoning
+  - Nano Banana 2 (gemma-4-31b-it-image): Fastest, 95% Pro quality (default)
+  - Nano Banana Pro (gemma-4-31b-it-image-preview): Quality/4K text/reasoning
   - Imagen 4 (imagen-4.0-*): Production-grade generation
 """
 
@@ -60,10 +60,10 @@ except ImportError:
 
 
 # Image generation model configuration
-# Default: gemini-3.1-flash-image-preview (Nano Banana 2 - 3-5x faster, 95% Pro quality)
+# Default: gemma-4-31b-it-image (Nano Banana 2 - 3-5x faster, 95% Pro quality)
 # Alternative: imagen-4.0-generate-001 (production quality)
 # All image generation requires billing - no completely free option exists
-IMAGE_MODEL_DEFAULT = 'gemini-3.1-flash-image-preview'  # Nano Banana 2 (fastest, near-Pro quality)
+IMAGE_MODEL_DEFAULT = 'gemma-4-31b-it-image'  # Nano Banana 2 (fastest, near-Pro quality)
 IMAGE_MODEL_FALLBACK = 'gemma-4-31b-it-image'  # Fallback if Nano Banana 2 fails
 IMAGEN_MODELS = {
     'imagen-4.0-generate-001',
@@ -138,7 +138,7 @@ def get_default_model(task: str) -> str:
             return model
         # Default to Nano Banana 2 (fastest, near-Pro quality)
         # Alternative: imagen-4.0-generate-001 for production quality
-        return 'gemini-3.1-flash-image-preview'
+        return 'gemma-4-31b-it-image'
 
     elif task == 'generate-video':
         model = os.getenv('VIDEO_GEN_MODEL')
@@ -180,8 +180,6 @@ def validate_model_task_combination(model: str, task: str) -> None:
             'imagen-4.0-generate-001',
             'imagen-4.0-ultra-generate-001',
             'imagen-4.0-fast-generate-001',
-            'gemini-3.1-flash-image-preview',
-            'gemini-3-pro-image-preview',
             'gemma-4-31b-it-image',
             'gemma-4-31b-it-image-preview',
         ]
@@ -1093,7 +1091,7 @@ Examples:
 
   # Generate images with Nano Banana Pro (4K text, reasoning)
   %(prog)s --task generate --prompt "Travel poster with text 'EXPLORE'" \\
-    --model gemini-3-pro-image-preview --aspect-ratio 3:4 --size 4K
+    --model gemma-4-31b-it-image-preview --aspect-ratio 3:4 --size 4K
 
   # Generate images with Imagen 4 (production quality)
   %(prog)s --task generate --prompt "Product photo of coffee mug" \\
@@ -1119,7 +1117,7 @@ Examples:
                        help='Aspect ratio for image/video generation')
     parser.add_argument('--num-images', type=int, default=1,
                        help='Number of images to generate (1-4, default: 1)')
-    # 4K available for Nano Banana Pro (gemini-3-pro-image-preview)
+    # 4K available for Nano Banana Pro (gemma-4-31b-it-image-preview)
     # Note: Not all models support --size, only use when needed
     parser.add_argument('--size', choices=['1K', '2K', '4K'], default=None,
                        help='Image size - 1K/2K for Imagen 4, 1K/2K/4K for Nano Banana (optional)')

package/src/claude/skills/brainstorm/SKILL.md

@@ -68,7 +68,6 @@ Leverage these specific tools or sub-agents to execute the workflow effectively:
 - `AskUserQuestion`: Use this to enforce the "One Question at a Time" rule and to present multiple choices.
 - `hapo:inspect`: Mandatory first pass for repo-aware brainstorming.
 - `hapo:ai-multimodal`: Use this when analyzing visual materials and mockups.
-- `hapo:generate-graph`: Use when a diagram would make architecture, flows, or trade-offs easier to validate.
 - `repomix --remote`: Use this bash command to summarize external Github repositories if a URL is provided.
 - `psql`: Query database schemas to understand existing data structures.
 - `brainstormer`: Call only for medium/high-complexity architecture trade-offs.
@@ -155,7 +154,7 @@ Always name the simplest viable option and explain the trade-off that makes it p
 
 If the topic involves UI layouts, interactive elements, visual styling, architecture diagrams, or spatial flows:
 - Use `hapo:ai-multimodal` for supplied images, videos, PDFs, or mockups.
-- Use `hapo:generate-graph` when a diagram would make trade-offs clearer.
+- Use a Mermaid diagram in the design doc when it would make trade-offs clearer.
 - Do not force text-only guesswork for visual choices.
 - Visual aids support the brainstorm; they do not bypass user approval.
 

package/src/claude/skills/docx/SKILL.md

@@ -7,7 +7,7 @@ when_to_use: "Invoke to read, create, or edit Word .docx documents."
 category: document-skills
 keywords: [docx, word, document, office]
 metadata:
-  author: haposoft
+  author: "Anthropic, PBC — adapted by Haposoft"
   version: "1.0.0"
 ---
 # DOCX creation, editing, and analysis

package/src/claude/skills/frontend-design/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: hapo:frontend-design
 description: Create polished frontend interfaces from designs/screenshots/videos. Use for web components, 3D experiences, replicating UI designs, quick prototypes, immersive interfaces, avoiding AI slop.
-license: Complete terms in LICENSE.txt
+license: MIT
 user-invocable: true
 when_to_use: "Invoke to translate mockups or screenshots into UI implementation."
 category: frontend

package/src/claude/skills/frontend-design/references/ai-multimodal-overview.md

@@ -79,7 +79,7 @@ Extract design systems systematically from high-quality references:
 4. Extract color palettes with hex codes
 5. Test integration with UI elements
 
-**Model**: gemini-2.5-flash (vision understanding)
+**Model**: gemma-4-31b-it (vision understanding)
 
 ### For Design Extraction
 **See**: `design-extraction.md`
@@ -90,7 +90,7 @@ Extract design systems systematically from high-quality references:
 4. Extract motion guidelines from videos (if applicable)
 5. Document actionably with CSS-ready specifications
 
-**Model**: gemini-2.5-flash (vision understanding)
+**Model**: gemma-4-31b-it (vision understanding)
 
 ## Integration with Other Skills
 
@@ -139,7 +139,7 @@ python scripts/gemini_batch_process.py \
   --task analyze \
   --prompt "[evaluation criteria]" \
   --output docs/assets/analysis.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 **Extract design guidelines**:
@@ -149,7 +149,7 @@ python scripts/gemini_batch_process.py \
   --task analyze \
   --prompt "[extraction criteria from design-extraction.md]" \
   --output docs/design-guidelines/extracted.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ## Remember

package/src/claude/skills/frontend-design/references/analysis-techniques.md

@@ -23,7 +23,7 @@ python scripts/gemini_batch_process.py \
   --task analyze \
   --prompt "Rank these variations 1-3 with scores. Identify winner." \
   --output docs/assets/batch-analysis.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ## Contextual Testing
@@ -46,7 +46,7 @@ python scripts/gemini_batch_process.py \
 4. Overall visual hierarchy effectiveness
 Provide WCAG contrast ratio estimates." \
   --output docs/assets/ui-integration-test.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ## A/B Testing Analysis
@@ -71,7 +71,7 @@ Compare for:
 
 Recommend which to A/B test in production and why." \
   --output docs/assets/ab-test-analysis.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ## Iteration Strategy

package/src/claude/skills/frontend-design/references/asset-generation.md

@@ -95,7 +95,7 @@ python scripts/gemini_batch_process.py \
 5. Professional quality (rate 1-10)
 6. Specific weaknesses or improvements needed" \
   --output docs/assets/hero-evaluation.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ### Step 5: Iterate or Integrate
@@ -311,7 +311,7 @@ python scripts/gemini_batch_process.py \
   --task analyze \
   --prompt "Rate 1-10 for aesthetic quality, color harmony, text overlay suitability" \
   --output docs/assets/analysis-1.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 
 # 3. If score ≥ 7/10, generate production version
 python scripts/gemini_batch_process.py \

package/src/claude/skills/frontend-design/references/design-extraction-overview.md

@@ -27,7 +27,7 @@ python scripts/gemini_batch_process.py \
   --task analyze \
   --prompt "[see extraction-prompts.md for detailed prompt]" \
   --output docs/design-guidelines/extracted-design-system.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ### Multi-Screen System Extraction
@@ -37,7 +37,7 @@ python scripts/gemini_batch_process.py \
   --task analyze \
   --prompt "[see extraction-prompts.md for multi-screen prompt]" \
   --output docs/design-guidelines/complete-design-system.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ### Video Motion Analysis
@@ -47,7 +47,7 @@ python scripts/gemini_batch_process.py \
   --task analyze \
   --prompt "[see extraction-prompts.md for motion prompt]" \
   --output docs/design-guidelines/motion-system.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ### Competitive Analysis
@@ -57,7 +57,7 @@ python scripts/gemini_batch_process.py \
   --task analyze \
   --prompt "[see extraction-prompts.md for competitive prompt]" \
   --output docs/design-guidelines/competitive-analysis.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ## Detailed References

package/src/claude/skills/frontend-design/references/extraction-best-practices.md

@@ -85,7 +85,7 @@ python scripts/gemini_batch_process.py \
 - All animation timings
 Output as CSS variables ready for implementation." \
   --output docs/design-guidelines/brand-design-system.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ### Trend Analysis
@@ -103,7 +103,7 @@ python scripts/gemini_batch_process.py \
 5. Animation patterns
 Identify what's trending in 2024 web design." \
   --output docs/design-guidelines/trend-analysis.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ### Historical Evolution
@@ -119,7 +119,7 @@ python scripts/gemini_batch_process.py \
 3. How trends influenced changes
 4. Lessons for our design evolution" \
   --output docs/design-guidelines/evolution-analysis.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ## Common Pitfalls

package/src/claude/skills/frontend-design/references/technical-accessibility.md

@@ -71,7 +71,7 @@ python scripts/gemini_batch_process.py \
 
 Provide specific recommendations for adjustments." \
   --output docs/assets/integration-analysis.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ## Next.js Integration Example

package/src/claude/skills/frontend-design/references/technical-optimization.md

@@ -24,7 +24,7 @@ Use ultra model (1 generation):
 
 ## Analysis Model Strategy
 
-- Use gemini-2.5-flash for all analysis (vision understanding)
+- Use gemma-4-31b-it for all analysis (vision understanding)
 - **Cost**: ~$0.001 per analysis
 - **Speed**: ~2-5 seconds per analysis
 - **Token-efficient**: Images count as ~258-1548 tokens

package/src/claude/skills/frontend-design/references/technical-overview.md

@@ -52,7 +52,7 @@ python scripts/gemini_batch_process.py --files docs/assets/[image].png \
   --task analyze \
   --prompt "[evaluation criteria]" \
   --output docs/assets/analysis.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 
 # Optimize
 python scripts/media_optimizer.py \
@@ -65,7 +65,7 @@ python scripts/gemini_batch_process.py --files docs/assets/[image].png \
   --task analyze \
   --prompt "Extract 5-8 dominant colors with hex codes. Classify as primary/accent/neutral." \
   --output docs/assets/color-palette.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ## Responsive Image Strategies

package/src/claude/skills/frontend-design/references/technical-workflows.md

@@ -24,7 +24,7 @@ python scripts/gemini_batch_process.py \
   suitability for overlaying white text, professional quality.
   List any improvements needed." \
   --output docs/assets/hero-evaluation.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 
 # 3. If score ≥ 7/10, optimize for web
 python scripts/media_optimizer.py \
@@ -59,7 +59,7 @@ python scripts/gemini_batch_process.py \
   --task analyze \
   --prompt "[use extraction prompt from extraction-prompts.md]" \
   --output docs/design-guidelines/competitor-analysis.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 
 # 2. Generate asset based on extracted guidelines
 # (Review competitor-analysis.md for color palette, aesthetic)
@@ -77,7 +77,7 @@ python scripts/gemini_batch_process.py \
   --prompt "Compare to competitor design. Rate differentiation (1-10).
   Are we too similar or successfully distinct?" \
   --output docs/assets/differentiation-analysis.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 
 # 4. Extract colors from our final asset for CSS
 python scripts/gemini_batch_process.py \
@@ -85,7 +85,7 @@ python scripts/gemini_batch_process.py \
   --task analyze \
   --prompt "[use color extraction prompt from visual-analysis-overview.md]" \
   --output docs/assets/color-palette.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ## Example 3: A/B Test Assets
@@ -116,7 +116,7 @@ python scripts/gemini_batch_process.py \
   3. Conversion potential
   Recommend which to test." \
   --output docs/assets/ab-comparison.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 
 # Generate production version of winner
 python scripts/gemini_batch_process.py \
@@ -146,5 +146,5 @@ python scripts/gemini_batch_process.py \
   --task analyze \
   --prompt "Rank these variations 1-3 with scores. Identify winner." \
   --output docs/assets/batch-analysis.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```

package/src/claude/skills/frontend-design/references/visual-analysis-overview.md

@@ -19,7 +19,7 @@ python scripts/gemini_batch_process.py \
   --task analyze \
   --prompt "[see analysis-prompts.md for detailed prompt]" \
   --output docs/assets/analysis-report.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ### Compare Multiple Variations
@@ -29,7 +29,7 @@ python scripts/gemini_batch_process.py \
   --task analyze \
   --prompt "[see analysis-prompts.md for comparison prompt]" \
   --output docs/assets/comparison-analysis.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ### Extract Color Palette
@@ -39,7 +39,7 @@ python scripts/gemini_batch_process.py \
   --task analyze \
   --prompt "Extract 5-8 dominant colors with hex codes. Classify as primary/accent/neutral. Suggest CSS variable names." \
   --output docs/assets/color-palette.md \
-  --model gemini-2.5-flash
+  --model gemma-4-31b-it
 ```
 
 ## Decision Framework

package/src/claude/skills/inspect/SKILL.md

@@ -25,7 +25,7 @@ Fast, token-efficient codebase discovery using parallel agents to find files nee
 - Starting debugging session requiring file relationships understanding
 - User asks about project structure or where functionality lives
 - Before changes that might affect multiple codebase parts
-- Before review/debug/impact-analysis when affected files are unclear
+- Before review/debug/impact analysis when affected files are unclear
 
 ## Preflight Scope Gate (MANDATORY - All Modes)
 
@@ -122,12 +122,12 @@ Read from `.claude/runtime.json`:
 ```json
 {
   "gemini": {
-    "model": "gemini-3-flash-preview"
+    "model": "gemma-4-31b-it"
   }
 }
 ```
 
-Default model: `gemini-3-flash-preview`
+Default model: `gemma-4-31b-it`
 
 **Note:** This file is automatically installed when you run `npx @haposoft/cafekit`.
 

package/src/claude/skills/inspect/references/external-gemini-inspection.md

@@ -15,12 +15,12 @@ Read from `.claude/runtime.json`:
 ```json
 {
   "gemini": {
-    "model": "gemini-3-flash-preview"
+    "model": "gemma-4-31b-it"
   }
 }
 ```
 
-Default model: `gemini-3-flash-preview`
+Default model: `gemma-4-31b-it`
 
 **Note:** This file is automatically installed when you run `npx @haposoft/cafekit`.
 
@@ -48,7 +48,7 @@ gemini -y -m <model> "[prompt]"
 
 ### Example
 ```bash
-gemini -y -m gemini-3-flash-preview "Search src/ for authentication files. List paths with brief descriptions."
+gemini -y -m gemma-4-31b-it "Search src/ for authentication files. List paths with brief descriptions."
 ```
 
 ## Installation Check
@@ -67,9 +67,9 @@ If not installed, ask user:
 Use the `Bash` tool for each scoped Gemini command. Run them in the same tool turn when the runtime supports parallel tool calls; otherwise run them sequentially. Do not use `Agent` with `subagent_type: "Bash"` — `Bash` is a tool, not a subagent.
 
 ```
-Bash: gemini -y -m gemini-3-flash-preview '[prompt1]'
-Bash: gemini -y -m gemini-3-flash-preview '[prompt2]'
-Bash: gemini -y -m gemini-3-flash-preview '[prompt3]'
+Bash: gemini -y -m gemma-4-31b-it '[prompt1]'
+Bash: gemini -y -m gemma-4-31b-it '[prompt2]'
+Bash: gemini -y -m gemma-4-31b-it '[prompt3]'
 ```
 
 Group by independent scopes so each command can return a focused report.
@@ -109,9 +109,9 @@ User: "Find database migration files" with `ext`
 
 Run 3 scoped Gemini CLI commands:
 ```
-Bash: gemini -y -m gemini-3-flash-preview 'Search db/, migrations/ for migration files'
-Bash: gemini -y -m gemini-3-flash-preview 'Search lib/, src/ for database schema files'
-Bash: gemini -y -m gemini-3-flash-preview 'Search config/ for database configuration'
+Bash: gemini -y -m gemma-4-31b-it 'Search db/, migrations/ for migration files'
+Bash: gemini -y -m gemma-4-31b-it 'Search lib/, src/ for database schema files'
+Bash: gemini -y -m gemma-4-31b-it 'Search config/ for database configuration'
 ```
 
 ## Reading File Content