@nockdev/awf 6.2.0 → 6.2.1

package/.agent/skills/ai-ml/gemini-live/data/code-execution.yaml

@@ -0,0 +1,131 @@
+# =============================================================================
+# code-execution.yaml — Gemini Code Execution Patterns
+# =============================================================================
+# 8 patterns for Python sandbox execution
+# Source: https://ai.google.dev/gemini-api/docs/code-execution
+# =============================================================================
+
+version: "1.0.0"
+total_patterns: 8
+
+overview:
+  description: "Model generates and runs Python code, iterates on results"
+  language: "Python only (can generate other languages, but only runs Python)"
+  cost: "No additional charge (standard token billing)"
+  sandbox: true
+  custom_libraries: false
+
+supported_libraries:
+  data_science:
+    - numpy
+    - pandas
+    - scipy
+    - scikit-learn
+  visualization:
+    - matplotlib
+    - seaborn
+  image:
+    - pillow
+    - opencv-python
+    - imageio
+  documents:
+    - PyPDF2
+    - python-docx
+    - python-pptx
+    - reportlab
+    - fpdf
+    - openpyxl
+    - xlrd
+  math:
+    - sympy
+    - mpmath
+  ml:
+    - tensorflow
+  parsing:
+    - lxml
+    - jsonschema
+    - pyparsing
+    - striprtf
+  other:
+    - chess
+    - geopandas
+    - jinja2
+    - joblib
+    - tabulate
+    - contourpy
+    - protobuf
+    - packaging
+
+api_usage:
+  enable:
+    python: |
+      from google import genai
+      from google.genai import types
+
+      client = genai.Client(api_key="API_KEY")
+      response = client.models.generate_content(
+          model="gemini-2.5-flash",
+          contents="Calculate the first 15 Fibonacci numbers and plot them",
+          config=types.GenerateContentConfig(
+              tools=[types.Tool(code_execution=types.ToolCodeExecution())]
+          )
+      )
+    javascript: |
+      const response = await ai.models.generateContent({
+          model: "gemini-2.5-flash",
+          contents: "Calculate Fibonacci and plot",
+          config: { tools: [{ codeExecution: {} }] }
+      });
+
+gemini3_image_execution:
+  description: "Gemini 3 Flash: write code to manipulate and inspect images"
+  use_cases:
+    zoom_and_inspect:
+      desc: "Model detects small details, crops and re-examines at higher resolution"
+      example: "Reading a distant gauge in a photo"
+    visual_math:
+      desc: "Multi-step calculations using code on image data"
+      example: "Summing line items on a receipt"
+    image_annotation:
+      desc: "Draw arrows, labels, highlights on images"
+      example: "Annotating a diagram to show relationships"
+  enable:
+    python: |
+      response = client.models.generate_content(
+          model="gemini-3-flash",
+          contents=[image, "What is the reading on the gauge?"],
+          config=types.GenerateContentConfig(
+              tools=[types.Tool(code_execution=types.ToolCodeExecution())]
+          )
+      )
+
+chat_mode:
+  description: "Use code execution in multi-turn chat"
+  benefit: "Model can build on previous code results"
+  pattern:
+    - "Turn 1: 'Load and analyze this data'"
+    - "Turn 2: 'Now visualize the top 10 results'"
+    - "Turn 3: 'Apply a trend line and predict next quarter'"
+
+billing:
+  description: "Code execution billing model"
+  rules:
+    - "Input tokens: billed once (your prompt)"
+    - "Intermediate tokens: generated code + execution results (billed as input)"
+    - "Output tokens: final summary + code + results"
+    - "Check usage_metadata for intermediate_token_count"
+
+limitations:
+  - "Can only generate and execute code (no media artifacts)"
+  - "Enabling may cause regressions in creative tasks"
+  - "Cannot install custom libraries"
+  - "Model variation in code execution quality"
+
+tool_combinations:
+  - "Code Execution + Google Search = Research + Analysis"
+  - "Code Execution + Function Calling = Compute + External Data"
+  - "All three combined via Live API"
+
+# =============================================================================
+# _DOMYH Awesome Code • Code Execution v1.0_
+# =============================================================================

package/.agent/skills/ai-ml/gemini-live/data/context-caching.yaml

@@ -0,0 +1,96 @@
+# =============================================================================
+# context-caching.yaml — Gemini Context Caching Patterns
+# =============================================================================
+# 6 patterns for implicit and explicit caching strategies
+# Source: https://ai.google.dev/gemini-api/docs/caching
+# =============================================================================
+
+version: "1.0.0"
+total_patterns: 6
+
+implicit_caching:
+  description: "Automatic cost savings — no code changes needed"
+  enabled: "Default (all Gemini models)"
+  effective_since: "May 8, 2025"
+  how_it_works:
+    - "Gemini auto-caches repeated input prefixes"
+    - "Cost savings passed on automatically"
+    - "No configuration needed"
+  optimization_tips:
+    - "Put large, common content at BEGINNING of prompt"
+    - "Send similar-prefix requests close together in time"
+    - "Check usage_metadata.cached_content_token_count for hits"
+  minimum_token_count:
+    gemini_2_5_flash: 1024
+    gemini_2_5_pro: 2048
+    gemini_3_flash: 1024
+
+explicit_caching:
+  description: "Manual cache creation with TTL control"
+  how_it_works:
+    - "Upload content to cache once"
+    - "Reference cached content in subsequent requests"
+    - "Set TTL (default: 1 hour)"
+    - "Pay reduced rate for cached tokens"
+
+  operations:
+    create:
+      python: |
+        from google import genai
+        client = genai.Client(api_key="API_KEY")
+        cache = client.caches.create(
+            model="gemini-2.5-flash",
+            config=genai.types.CreateCachedContentConfig(
+                display_name="my-cache",
+                contents=[large_document],
+                ttl="3600s"  # 1 hour
+            )
+        )
+
+    use:
+      python: |
+        response = client.models.generate_content(
+            model="gemini-2.5-flash",
+            contents="Summarize the key findings",
+            config=genai.types.GenerateContentConfig(
+                cached_content=cache.name
+            )
+        )
+
+    list:
+      python: |
+        for cache in client.caches.list():
+            print(f"{cache.display_name}: {cache.expire_time}")
+
+    update_ttl:
+      python: |
+        client.caches.update(
+            name=cache.name,
+            config=genai.types.UpdateCachedContentConfig(ttl="7200s")
+        )
+
+    delete:
+      python: |
+        client.caches.delete(name=cache.name)
+
+use_cases:
+  - scenario: "Chatbot with large system instructions"
+    benefit: "Cache system prompt, pay once"
+  - scenario: "Repetitive video analysis"
+    benefit: "Cache video, query multiple times"
+  - scenario: "Document Q&A"
+    benefit: "Cache document set, ask many questions"
+  - scenario: "Code repository analysis"
+    benefit: "Cache codebase, run multiple audits"
+
+cost_analysis:
+  description: "Cost savings from explicit caching"
+  factors:
+    - "Cached token rate < regular input rate"
+    - "Savings increase with more requests to same cache"
+    - "TTL cost: longer cache = higher storage cost"
+    - "Break-even: typically 3-5 requests to same content"
+
+# =============================================================================
+# _DOMYH Awesome Code • Context Caching v1.0_
+# =============================================================================

package/.agent/skills/ai-ml/gemini-live/data/grounding.yaml

@@ -0,0 +1,97 @@
+# =============================================================================
+# grounding.yaml — Grounding with Google Search Patterns
+# =============================================================================
+# 6 patterns for real-time web grounding and citation building
+# Source: https://ai.google.dev/gemini-api/docs/grounding
+# =============================================================================
+
+version: "1.0.0"
+total_patterns: 6
+
+overview:
+  description: "Connect Gemini to real-time web content via Google Search"
+  benefits:
+    - "Increase factual accuracy (reduce hallucinations)"
+    - "Access real-time information"
+    - "Provide verifiable citations"
+  supported_languages: "All available languages"
+
+how_it_works:
+  flow:
+    - step: "User Prompt"
+      desc: "Your app sends prompt with google_search tool enabled"
+    - step: "Prompt Analysis"
+      desc: "Model determines if search can improve answer"
+    - step: "Google Search"
+      desc: "Model auto-generates and executes search queries"
+    - step: "Results Processing"
+      desc: "Model synthesizes search results into response"
+    - step: "Grounded Response"
+      desc: "Response includes text + groundingMetadata with citations"
+
+api_usage:
+  enable:
+    python: |
+      from google import genai
+      from google.genai import types
+
+      client = genai.Client(api_key="API_KEY")
+      response = client.models.generate_content(
+          model="gemini-2.5-flash",
+          contents="Who won Euro 2024?",
+          config=types.GenerateContentConfig(
+              tools=[types.Tool(google_search=types.GoogleSearch())]
+          )
+      )
+    javascript: |
+      const response = await ai.models.generateContent({
+          model: "gemini-2.5-flash",
+          contents: "Who won Euro 2024?",
+          config: { tools: [{ googleSearch: {} }] }
+      });
+
+response_structure:
+  groundingMetadata:
+    webSearchQueries:
+      desc: "Array of search queries used"
+      use: "Debugging, understanding model reasoning"
+    searchEntryPoint:
+      desc: "HTML/CSS for search suggestion widget"
+      use: "Required by Terms of Service for web display"
+    groundingChunks:
+      desc: "Array of web sources (uri + title)"
+      use: "Build source list, link verification"
+    groundingSupports:
+      desc: "Maps text segments to source citations"
+      fields:
+        - "segment.startIndex / endIndex — text range"
+        - "segment.text — the grounded text"
+        - "groundingChunkIndices — which sources support this claim"
+      use: "Build inline citations, highlight sourced claims"
+
+building_citations:
+  description: "How to create inline citations from grounding response"
+  steps:
+    - "Parse groundingChunks for source URLs"
+    - "Parse groundingSupports for text-to-source mapping"
+    - "For each support, link text range to source(s)"
+    - "Render as [text](source_url) or footnotes"
+  example: |
+    # Text: "Spain won Euro 2024, defeating England 2-1"
+    # groundingSupports[0].groundingChunkIndices = [0]
+    # groundingChunks[0].web.uri = "https://..."
+    # → "Spain won Euro 2024 [1], defeating England 2-1"
+
+url_context_tool:
+  description: "Combine grounding with specific URL analysis"
+  how: "Use URL context tool alongside Google Search"
+  benefit: "Ground in both public web AND specific URLs you provide"
+
+pricing:
+  description: "Grounding incurs search API costs"
+  tip: "Use grounding selectively — not all queries need real-time data"
+  free_tier: "Available with API key"
+
+# =============================================================================
+# _DOMYH Awesome Code • Grounding v1.0_
+# =============================================================================

package/.agent/skills/ai-ml/gemini-live/data/live-api.yaml

@@ -0,0 +1,103 @@
+# =============================================================================
+# live-api.yaml — Gemini Live API Patterns
+# =============================================================================
+# 8 patterns for real-time WebSocket streaming with Gemini
+# Source: https://ai.google.dev/gemini-api/docs/live
+# =============================================================================
+
+version: "1.0.0"
+total_patterns: 8
+
+overview:
+  description: "Low-latency, real-time voice and video interactions with Gemini"
+  transport: "WebSocket"
+  input_formats:
+    audio: "PCM 16-bit, 16kHz, mono"
+    video: "Supported camera streams"
+    text: "UTF-8 text"
+  output_audio: "24kHz sample rate"
+  features:
+    - "Voice Activity Detection (VAD)"
+    - "Tool use and function calling"
+    - "Session management"
+    - "Ephemeral tokens"
+    - "Multi-tool support"
+
+approaches:
+  server_to_server:
+    desc: "Backend proxies client streams to Live API"
+    pros:
+      - "Full security control"
+      - "API key stays server-side"
+      - "Can add business logic"
+    cons:
+      - "Added latency from proxy"
+      - "Server resource requirements"
+    use_case: "Production apps with security requirements"
+
+  client_to_server:
+    desc: "Frontend connects directly to Live API"
+    pros:
+      - "Lowest latency"
+      - "Simpler architecture"
+    cons:
+      - "Requires ephemeral tokens"
+      - "Limited security control"
+    use_case: "Real-time demos, low-latency requirements"
+
+voice_activity_detection:
+  desc: "Automatic detection of when user starts/stops speaking"
+  behavior: "Model processes accumulated audio when silence detected"
+  customization: "Configurable silence threshold"
+
+session_management:
+  desc: "Maintain conversation state across multiple exchanges"
+  features:
+    - "Session persistence"
+    - "Context window management"
+    - "Automatic reconnection"
+  max_duration: "Configurable (model dependent)"
+
+ephemeral_tokens:
+  desc: "Short-lived tokens for client-side auth"
+  flow:
+    - "Server generates ephemeral token from API key"
+    - "Token sent to client"
+    - "Client uses token to connect directly"
+    - "Token expires after set duration"
+  security: "Never expose API key to client"
+
+partner_integrations:
+  - name: "Pipecat (Daily)"
+    desc: "Real-time AI chatbot framework"
+    url: "https://docs.pipecat.ai/guides/features/gemini-live"
+  - name: "LiveKit"
+    desc: "Agents with Gemini Live"
+    url: "https://docs.livekit.io/agents/models/realtime/plugins/gemini/"
+  - name: "ADK (Google)"
+    desc: "Agent Development Kit with Live API"
+    url: "https://google.github.io/adk-docs/streaming/"
+  - name: "Fishjam"
+    desc: "Live video/audio streaming"
+    url: "https://docs.fishjam.io/tutorials/gemini-live-integration"
+  - name: "Voximplant"
+    desc: "Phone call integration"
+    url: "https://voximplant.com/products/gemini-client"
+
+api_usage:
+  python: |
+    from google import genai
+    client = genai.Client(api_key="API_KEY")
+
+    async with client.aio.live.connect(model="gemini-2.0-flash-live-001") as session:
+        # Send audio
+        await session.send(audio_data, end_of_turn=True)
+        
+        # Receive response
+        async for response in session.receive():
+            if response.data:
+                play_audio(response.data)
+
+# =============================================================================
+# _DOMYH Awesome Code • Live API v1.0_
+# =============================================================================

package/.agent/skills/ai-ml/gemini-media-gen/META.yaml

@@ -0,0 +1,56 @@
+name: gemini-media-gen
+version: "1.0.0"
+display: Gemini Media Generation
+category: ai-ml
+tier: 1
+priority: 2
+
+desc: "Gemini native media generation: Nano Banana image gen (Flash + Pro 4K), Veo 3.1 video gen (8s 720p-4K + audio), Files API integration. Prompt optimization, multi-ref images, search grounding, batch generation. 36+ patterns."
+
+triggers:
+  file_patterns: []
+  keywords:
+    - nano banana
+    - image generation
+    - generate image
+    - veo
+    - video generation
+    - generate video
+    - gemini image
+    - gemini video
+    - text to image
+    - text to video
+    - image editing
+    - 4k image
+    - search grounding image
+    - files api
+    - media generation
+    - aspect ratio
+    - image prompt
+  intents:
+    - generate image
+    - create video
+    - edit image
+    - media generation
+    - batch image creation
+
+caps:
+  - Nano Banana Flash (speed, 1024px max)
+  - Nano Banana Pro (4K, Thinking, 14 ref images, Search grounding)
+  - Veo 3.1 (8s video, 720p-4K, native audio)
+  - Video extension & frame control
+  - Files API (20GB/project, 48h retention, free)
+  - Batch image generation
+  - Multi-reference image compositing
+  - Text rendering in images
+  - Image editing (inpainting, style transfer)
+
+related_skills:
+  - prompt-engineering
+  - domyh-design
+
+data_files:
+  - data/image-models.yaml
+  - data/image-prompts.yaml
+  - data/video-generation.yaml
+  - data/files-api.yaml

package/.agent/skills/ai-ml/gemini-media-gen/SKILL.md

@@ -0,0 +1,128 @@
+# Gemini Media Generation
+
+> 🎨 **Native image + video generation via Gemini API**
+> **Models**: Nano Banana (Flash/Pro) + Veo 3.1 | **Patterns**: 36+
+
+---
+
+## Quick Reference
+
+| What You Need                | Data File               | Patterns |
+| ---------------------------- | ----------------------- | -------- |
+| Model selection & configs    | `image-models.yaml`     | 8        |
+| Image prompt optimization    | `image-prompts.yaml`    | 12       |
+| Video generation & prompting | `video-generation.yaml` | 10       |
+| Files API management         | `files-api.yaml`        | 6        |
+
+---
+
+## Model Selection Guide
+
+| Feature              | Nano Banana Flash                           | Nano Banana Pro                           |
+| -------------------- | ------------------------------------------- | ----------------------------------------- |
+| **Speed**            | ⚡ Fast                                     | 🐢 Slower                                 |
+| **Max resolution**   | 1024×1024                                   | **4K (4096×4096)**                        |
+| **Reference images** | Up to 5                                     | **Up to 14**                              |
+| **Thinking**         | ❌                                          | ✅                                        |
+| **Search grounding** | ❌                                          | ✅                                        |
+| **Text rendering**   | Basic                                       | **Enhanced**                              |
+| **SynthID**          | ✅ (watermark)                              | ✅                                        |
+| **Model ID**         | `gemini-2.0-flash-preview-image-generation` | `gemini-2.5-pro-preview-image-generation` |
+
+**Decision**: Use Flash for speed/prototyping, Pro for quality/production.
+
+---
+
+## Image Generation Rules
+
+1. **Describe the scene** — Don't list keywords
+   - ✅ "A golden retriever playing fetch on a sandy beach at sunset, waves crashing in the background"
+   - ❌ "golden retriever, beach, sunset, waves, fetch, sand"
+
+2. **Aspect Ratios**: `1:1` (default), `16:9` (landscape), `9:16` (portrait), `4:3`, `3:4`
+
+3. **Safety**: All images include invisible SynthID watermark. Safety filters block harmful/CSAM content.
+
+4. **Text in images**: Wrap desired text in quotes — `'Sale 50% Off'`
+
+5. **Vietnamese support**: Gemini supports Vietnamese text in prompts natively
+
+---
+
+## Reference Image Strategies
+
+| Strategy                  | Max Refs          | Use Case                                        |
+| ------------------------- | ----------------- | ----------------------------------------------- |
+| **Style transfer**        | 1-3               | Apply reference art style to new content        |
+| **Subject reference**     | 1                 | Keep consistent character/product across images |
+| **Multi-ref composition** | Up to 14 (Pro)    | Combine elements from multiple references       |
+| **Inpainting**            | 1 (mask required) | Edit specific regions of existing image         |
+
+---
+
+## Veo 3.1 Video Generation
+
+| Parameter           | Option          | Description                           |
+| ------------------- | --------------- | ------------------------------------- |
+| **Duration**        | 5-8 seconds     | Fixed short clips                     |
+| **Resolution**      | 720p, 1080p, 4K | Output quality                        |
+| **Audio**           | Native          | AI-generated sound effects & dialogue |
+| **Input**           | Text or Image   | Text-to-video or image-to-video       |
+| **Negative prompt** | Yes             | "Don't include X in the video"        |
+
+### Video Prompt Elements
+
+```
+Subject: [who/what is in the scene]
+Action: [what is happening]
+Style: [cinematic, documentary, anime, etc.]
+Camera: [pan left, zoom in, tracking shot, etc.]
+Audio: [ambient sounds, dialogue, music style]
+Lighting: [golden hour, neon, soft diffused, etc.]
+```
+
+### Video Extension
+
+Feed last frame of generated video to extend the scene for another 5-8 seconds.
+
+### Frame Control
+
+Use reference image as first/last frame to control video composition.
+
+---
+
+## Files API Quick Reference
+
+| Feature             | Value                            |
+| ------------------- | -------------------------------- |
+| **Storage**         | 20GB per project                 |
+| **Per-file max**    | 2GB                              |
+| **Retention**       | 48 hours                         |
+| **Cost**            | Free                             |
+| **Supported types** | Images, video, audio, PDFs, text |
+
+```python
+# Upload
+file = client.files.upload(file="large_video.mp4")
+
+# Check status (for video processing)
+while file.state == "PROCESSING":
+    file = client.files.get(name=file.name)
+
+# Use in generation
+response = model.generate_content([file, "Describe this video"])
+
+# Delete
+client.files.delete(name=file.name)
+```
+
+---
+
+## HSA Integration
+
+| Domain | Query Examples                          |
+| ------ | --------------------------------------- |
+| Image  | "nano banana 4K pro reference images"   |
+| Video  | "veo 3.1 camera movement prompt"        |
+| Files  | "files API upload video processing"     |
+| Prompt | "image generation describe scene style" |