메뉴
Building voice and video agents with Gemini Live API — bidirectional streaming via WebSocket, LiveRequestQueue, run_live(), RunConfig, and audio/video patterns. Load when the user wants a voice agent, live agent, or real-time audio/video capabilities.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Agent architecture patterns for Google ADK — when and how to use LlmAgent, LoopAgent, SequentialAgent, ParallelAgent, and multi-agent hierarchies. Load this skill when deciding the agent's architecture.
ADK callback signatures, HITL approval gates, and state management — exact parameter names, before/after hooks, EventActions.state_delta, and defensive HITL patterns. Load this skill when adding callbacks or approval gates to an agent.
Writing effective ADK system prompts — dynamic instructions with InstructionProvider, state placeholders, context injection, tone/style sections. Load this skill when creating or improving an agent's system prompt.
Creating valid SKILL.md files following the agentskills.io specification — frontmatter, instructions, references directory, progressive disclosure (L1/L2/L3), and SkillToolset wiring in agent.py. Load this skill when generating domain skills for an agent.
Five canonical skill design patterns for ADK agents: Tool Wrapper, Generator, Reviewer, Inversion, and Pipeline. Load this skill during Step 2 (Design) to recommend which patterns fit the user's agent, and during Step 4 (Generate) to get skeleton templates for each pattern.
Building Google ADK function tools — ToolContext usage, proper signatures, error handling, return formats, type hints, and async patterns. Load this skill when creating custom tools for an agent.
| name | adk-streaming |
| description | Building voice and video agents with Gemini Live API — bidirectional streaming via WebSocket, LiveRequestQueue, run_live(), RunConfig, and audio/video patterns. Load when the user wants a voice agent, live agent, or real-time audio/video capabilities. |
Build real-time voice and video agents using ADK's bidirectional streaming with the Gemini Live API.
The user explicitly asks for:
Browser (mic/cam) → WebSocket /ws/{user_id}/{session_id}
↓ upstream_task
LiveRequestQueue (text: send_content, audio: send_realtime)
↓
Runner.run_live() ←→ Gemini Live API (BIDI mode)
↓ downstream_task
WebSocket → Browser (audio playback + transcriptions)
| Component | Purpose |
|---|---|
LiveRequestQueue | Async FIFO buffer between WebSocket and agent. One per session. |
Runner.run_live() | Async generator consuming from queue, yielding events |
RunConfig(streaming_mode=StreamingMode.BIDI) | Configures BIDI mode for Live API |
StreamingMode.BIDI | WebSocket to Gemini Live API (audio/video) |
StreamingMode.SSE | HTTP streaming to standard Gemini API (text only) |
The generated template includes streaming support gated behind STREAMING_ENABLED=true:
streaming.py — WebSocket handler with upstream/downstream tasksrun_adk.py — conditional branch that builds Runner manually with LIVE_MODELstatic/test_client.html — minimal browser test clientconfig/llm.py — LIVE_MODEL config (Gemini model string, not LiteLLM)When designing a streaming agent:
LIVE_MODEL), not OpenRouter/LiteLLMGOOGLE_API_KEY (not OPENROUTER_API_KEY)When generating a streaming agent, customize these in streaming.py:
speech_config to RunConfig for voice selectionlanguage_code in speech configresponse_modalitiesLoad references for details on each customization.
streaming-patterns for RunConfig options, VAD modes, voice selection, and audio format details.live-api-reference for LiveRequestQueue API, event types, and error handling.