google-gemini/gemini-skills
4 skills22.4k installs14.4k starsGitHub
Install
npx skills add https://github.com/google-gemini/gemini-skillsSkills in this repo
1Gemini Api Devgemini-api-dev is an agent skill for solo builders shipping features on Google’s hosted Gemini and Gemma models. It exists because default model knowledge in agents goes stale quickly: the skill enforces current model names (for example gemini-3.5-flash, gemini-3.1-pro-preview, gemma-4 variants) and warns against deprecated gemini-2.0 and gemini-1.5 families. Use it when you are choosing SDKs across Python, JavaScript/TypeScript, Java, or Go, sending multimodal payloads, enabling function calling, or requiring structured JSON-style outputs. It emphasizes API capabilities and selection tradeoffs—speed versus reasoning, cost-efficient flash-lite versus pro, image preview models for generation/editing—rather than marketing copy. The pattern is integration reference plus procedural guardrails so codegen matches live API behavior. Pair it with your app’s backend or agent orchestration layer; prerequisites are a Google AI or Vertex setup consistent with your deployment choice and API keys managed outside the skill.13.3kinstalls2Gemini Interactions ApiGemini Interactions API is a migration-oriented agent skill for developers who already ship against the classic Gemini API and need to move to the Interactions API and newer model families without breaking production behavior. The documented workflow starts with scope discipline: if the user did not name a single file, directory, or explicit list, the agent must ask whether to migrate the whole repo, a subtree, or specific files, and may use ripgrep counts per directory to make that choice concrete. Only unambiguous scope (or a confirmed answer) unlocks edits. The skill references Google’s full migration guide for detailed replacements across text generation, conversation, streaming, tools, structured output, grounding, and multimodal inputs. Solo builders embedding Gemini in agents, APIs, or internal tools use it during Build when integrations are the bottleneck, especially when legacy symbol names like generateContent or older model IDs still litter the tree.4.2kinstalls3Gemini Live Api DevGemini Live API Dev is an agent skill for solo builders shipping conversational or multimodal experiences that cannot wait on REST round-trips. It walks you through WebSocket-based bidirectional streaming with the Gemini Live API: microphone-to-speaker audio, camera or screen video frames alongside audio, inline text, transcriptions on both sides, and automatic interruption via Voice Activity Detection. You get practical coverage of native audio behavior, synchronous function calling within a live session, real-time Google Search grounding, and the configuration surface teams usually discover only after production failures—session compression, resumption, and GoAway handling. The skill explicitly targets the supported client stacks (google-genai in Python and @google/genai in TypeScript/JavaScript) and ephemeral token flows so browser or mobile clients authenticate without embedding secrets. Use it when you are past a static chat demo and need a dependable streaming architecture before launch.3.7kinstalls4Vertex Ai Api Devvertex-ai-api-dev steers coding agents when a builder asks for Gemini in an enterprise setting or names Vertex AI explicitly. It centers on Google’s unified Gen AI SDK across Python, JavaScript/TypeScript, Go, Java, and C#, and it forbids drifting to deprecated clients such as `google-cloud-aiplatform`, `@google-cloud/vertexai`, or standalone `google-generativeai` where the skill directs otherwise. Capability coverage spans everyday text generation through function calling, schema-bound structured output, embeddings, context caching for large prompts, batch prediction for offline datasets, and the Live API for streaming voice and video interactions. Compatibility notes require active Google Cloud credentials and an enabled Vertex AI API, which shapes how solo builders prototype locally versus in a billed project. Use during Build while you integrate model calls into backends, workers, or agent tools—not when you are only comparing models at the idea stage. The skill is opinionated infrastructure guidance meant to keep production paths consistent with current Google documentation.1.3kinstalls