Gemini LiveAPI Client Service Skill
Generate a resumable, multimodal, bidirectional Gemini LiveAPI client over WebSockets.
The evidence specifies ADC bearer authentication, isolated environments, no system-Python mutation, and avoidance of long-lived API keys, while describing the WebSocket data flow. It does not adequately address sensitive audio/video/transcript handling, user confirmation, deployment privilege boundaries, dependency security, rollback, or recovery after failure, so 7 points are deducted.
The steps, APIs, reconnection behavior, resumption logic, and abnormal paths are described in useful detail, with a validation checklist. However, the selected skill contains no committed implementation, test suite, or reproducible CI evidence; reconnection indexing and cross-language details remain for the generator to implement, so the score is limited by static calibration and further reduced for uncertainty.
The scope clearly targets Gemini Enterprise LiveAPI bidirectional multimodal WebSocket clients and explicitly excludes ordinary non-live Gemini API use. Language boundaries, trigger precision, non-fit environments, Chinese-language guidance, and mainland-China reachability are insufficiently documented; the core function depends on Google Cloud services that may be difficult to reach from mainland China, so points are deducted.
The material includes prerequisites, reference files, staged instructions, required outputs, a validation checklist, repository-level Apache-2.0 licensing, and an active-development note. It lacks skill-specific versioning, changelog, named maintenance responsibility or update path, pinned dependency guidance, and troubleshooting FAQs, so 7 points are deducted.
The requested outputs are concrete and broad: client, tests, run documentation, UI, backend, multimodal handling, transcription, interruption, and resumption. Static files do not establish that generated artifacts will run directly; the demo scope is substantial and requires user-supplied project and media details, so review and debugging are likely, warranting deductions.
Protocol and session-management references plus official documentation links make key requirements auditable. There is no implementation, committed test coverage, CI evidence, or third-party execution result for this skill, so the conclusion relies mainly on author-written specifications and receives only limited credit.
- This is a low-confidence static review; the generation flow and tests were not executed.
- Before generation, confirm user authorization for deployment scripts and frontend/backend services, and define handling, retention, and transmission of audio, video, transcripts, and credentials.
- Validate the current official LiveAPI protocol, model, voice, and language support; the referenced material may become stale or contain field/naming inconsistencies.
- Mainland-China users may not have stable access to the core Google Cloud services; reachability or alternatives should be documented.
- Dependencies should be isolated, version-pinned, and source-reviewed; failure, expired credentials or session handles, and rollback paths need explicit treatment.
What does this skill do, and when should you use it?
This skill generates a client service class in the user’s chosen programming language for the Gemini Enterprise LiveAPI WebSocket endpoint. The generated client handles ADC bearer-token authentication, session setup, transparent session resumption, and ClientMessage/ServerMessage proto exchange. Python is the default and reference language, with an isolated environment required inside the output folder when used. The skill also generates a frontend and backend demo for validating text, audio, video, transcription, and interruption behavior.
It copies the message Markdown, proto definition, and session-management references into a user-provided output folder, reconciles them with linked public documentation when needed, and generates the client class, test file, dependencies, isolated-runtime setup script, how_to_run.md, frontend UI, backend service, and how_to_test_with_ui.md. The client connects to the specified Google Cloud Live API WebSocket, obtains and refreshes bearer tokens through ADC, sends setup, realtime input, and client content, and yields parsed ServerMessage instances. The demo UI supports opening or closing connections, selecting a model, streaming microphone, camera, or screen input, sending text, and displaying model audio and both-party transcription.
- A developer needs a Gemini Enterprise LiveAPI WebSocket client for a Google Cloud project.
- An application team needs to stream text, audio, or video and receive model responses incrementally.
- An engineer needs session recovery after network disconnects, go_away messages, or specified WebSocket closures.
- A prototype team needs a browser demo to validate audio playback, video input, transcription, and interruption handling.
What are this skill's strengths and limitations?
- Covers WebSocket bidirectional streaming, ADC authentication, token refresh, and transparent session resumption.
- Specifies async APIs, local proto types, isolated Python environments, and reconnect buffering behavior.
- Generates client tests, a frontend UI, a backend service, and usage documentation.
- Defines UI behavior for interruption and transcription finished signals in detail.
- Requires a Google Cloud project, enabled APIs, and ADC configuration.
- Python is the default and reference language, while implementation details for other languages are not expanded in the source.
- The generation flow requires checking public documentation and therefore needs network access.
- The source provides no test results, supported-platform matrix, or cost information.
How do you install this skill?
In an environment that supports Agent Skill installation, run: npx skills add google/skills. Select the skill corresponding to skills/cloud/gemini-live-api/SKILL.md during installation. The source does not specify a dedicated client invocation command or platform-specific installation procedure.
How do you use this skill?
Prepare a Google Cloud project with the Vertex AI / Gemini Enterprise Agent Platform APIs enabled, then run on the host that will execute the client: gcloud auth application-default login. Ask for a request such as “Generate a Python Gemini Enterprise LiveAPI client service for my project, write it to /tmp/liveapi_out, and include text, audio, video tests, and a browser demo.” Provide project_id, location, model_id, implementation language, media samples, and any information needed to run the test. The source does not define a universal trigger syntax.
How does this skill compare with similar options?
It explicitly excludes one-shot generateContent, embeddings, image or video generation, and fine-tuning; those non-live Gemini API scenarios should use the repository’s gemini-api skill.