Skip to main content الرئيسية المنشئون amo-tech-ai rocket-path-ai gemini
gemini Use when integrating Gemini AI models in Supabase Edge Functions or Deno. Covers Gemini 3 models (Pro, Flash, Pro-Image), thinking levels, URL Context, Google Search grounding, structured output, thought signatures, function calling, image generation, document processing, deep search, error handling, and REST API patterns. Use for AI enrichment, data extraction, analysis tasks, and image generation.
الانتقال إلى التثبيت سوق المهارات اكتشف واستكشف مهارات الذكاء الاصطناعي التي بناها المجتمع.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
عرض تفاصيل Prompt يتجاوز الأمر المباشر Prompt المخصّص للمراجعة. افحص المصدر قبل تشغيله.
npx skills add https://github.com/amo-tech-ai/rocket-path-ai --skill geminiيبقى الأمر في سطر واحد. مرّر أفقيًا لمراجعته كاملًا قبل النسخ.
تفضّل نسخة محلية؟ نزّل الملفات المتاحة حاليًا لدى SkillsMP.
تحميل Zip جاري التحميل... المزيد من هذا المستودع OpenClaw Gateway setup, configuration, and best practices. Use when installing OpenClaw, configuring channels (WhatsApp/Telegram/Discord), setting up providers (OpenAI/Google/Anthropic), creating skills, managing the gateway daemon, or troubleshooting OpenClaw issues. Triggers on: openclaw, gateway, whatsapp channel, telegram bot, openclaw skill, openclaw config, openclaw install, openclaw cron, openclaw security.
Use when the user asks how to build with OpenAI products or APIs and needs up-to-date official documentation with citations, help choosing the latest model for a use case, or explicit GPT-5.4 upgrade and prompt-upgrade guidance; prioritize OpenAI docs MCP tools, use bundled references only as helper context, and restrict any fallback browsing to official OpenAI domains.
Use this skill for setting up vector similarity search with pgvector for AI/ML embeddings, RAG applications, or semantic search.
**Trigger when user asks to:**
- Store or search vector embeddings in PostgreSQL
- Set up semantic search, similarity search, or nearest neighbor search
- Create HNSW or IVFFlat indexes for vectors
- Implement RAG (Retrieval Augmented Generation) with PostgreSQL
- Optimize pgvector performance, recall, or memory usage
- Use binary quantization for large vector datasets
**Keywords:** pgvector, embeddings, semantic search, vector similarity, HNSW, IVFFlat, halfvec, cosine distance, nearest neighbor, RAG, LLM, AI search
Covers: halfvec storage, HNSW index configuration (m, ef_construction, ef_search), quantization strategies, filtered search, bulk loading, and performance tuning.
**This project:** We use OpenAI text-embedding-3-small (1536) and store as vector(1536) in knowledge_chunks. halfvec is an optional future optimization; apply this skill's tuning (ef_search, iterative_scan
المهن ذات الصلة SOC
استنادا إلى تصنيف SOC المهني
name gemini description Use when integrating Gemini AI models in Supabase Edge Functions or Deno. Covers Gemini 3 models (Pro, Flash, Pro-Image), thinking levels, URL Context, Google Search grounding, structured output, thought signatures, function calling, image generation, document processing, deep search, error handling, and REST API patterns. Use for AI enrichment, data extraction, analysis tasks, and image generation.
Gemini AI Integration
Build AI features using Google's Gemini 3 models with URL Context, Google Search grounding, structured outputs, thinking levels, function calling, and image generation.
CRITICAL RULES (Audit Findings G1-G6)
These rules MUST be followed in every Gemini integration. Violating them causes crashes, degraded output, or security issues.
ID Rule Why G1 Always use responseJsonSchema with responseMimeType: "application/json" responseMimeType alone just asks the model to try JSON. Adding responseJsonSchema guarantees syntactically valid JSON matching the schema. Without it, JSON parse crashes are likely.G2 Keep temperature at default 1.0 for Gemini 3 Setting temperature below 1.0 causes looping, repetitive output, or degraded performance on Gemini 3 models. The docs explicitly warn against this. G3 Combine structured output with Google Search Gemini 3 supports responseJsonSchema + googleSearch together. Use both to get valid JSON even with search grounding. Without the schema, search metadata can corrupt JSON output.
G4 Pass API key via x-goog-api-key header, NOT query param ?key= exposes the key in server logs and referrer headers. Every official REST example uses the header.
G5 Extract groundingChunks citations from Google Search The response includes groundingChunks (source URLs/titles) and groundingSupports (inline citation mapping). Don't discard this data — use it for citations.
G6 Both googleSearch and google_search casing work The API accepts both camelCase and snake_case via protobuf. Use camelCase for SDK consistency.
Quick Reference Model Use Case Context Pricing gemini-3.1-pro-previewLatest Pro — complex reasoning, agentic, vibe coding 1M / 64k $2-4 / $12-18 gemini-3-flash-previewFast responses, extraction, chat 1M / 64k $0.50 / $3 gemini-3.1-flash-lite-previewCost-efficient, fastest for high-frequency tasks 1M / 64k $0.50 / $3 gemini-3.1-flash-image-previewFast image generation + editing 131k / 32k $2 / $0.134 gemini-3-pro-image-previewPro image generation (Nano Banana Pro) 131k / 32k $2 / $0.134
⚠️ DEPRECATED: gemini-3-pro-preview shut down March 9, 2026 — use gemini-3.1-pro-preview instead.
All Gemini 3 models have a knowledge cutoff of January 2025 .
SDK Setup (Deno Edge Functions) import { GoogleGenAI } from "npm:@google/genai@^1.0.0" ;
const ai = new GoogleGenAI ({ apiKey : Deno .env .get ("GEMINI_API_KEY" ) });
IMPORTANT: Use npm:@google/genai@^1.0.0 (not esm.sh, not legacy @google/generativeai).
Thinking Levels Control reasoning depth with thinkingLevel (Gemini 3):
Level Models Use Case Latency minimalFlash only Matches "no thinking" for most queries Lowest lowPro + Flash Simple instruction following Low mediumFlash only Balanced tasks Medium highPro + Flash Complex analysis, scoring (default) Higher
const response = await ai.models .generateContent ({
model : "gemini-3-flash-preview" ,
contents : prompt,
config : {
thinkingConfig : {
thinkingLevel : "high"
}
}
});
Best Practice: Use high for scoring, readiness calculations, and complex analysis. Use low or minimal for simple extraction and chat.
WARNING: Cannot use both thinkingLevel and legacy thinkingBudget in the same request (400 error).
Built-in Tools
1. URL Context Extract data from web URLs directly:
const response = await ai.models .generateContent ({
model : "gemini-3-flash-preview" ,
contents : `Extract company info from ${url} ` ,
config : {
tools : [{ urlContext : {} }]
}
});
Up to 20 URLs per request
Max 34MB per URL
Cannot access: paywalled content, YouTube, Google Workspace, localhost
Supported: HTML, JSON, PDF, images (PNG, JPEG)
const metadata = response.candidates [0 ].urlContextMetadata ;
2. Google Search Grounding Connect to real-time web information:
const response = await ai.models .generateContent ({
model : "gemini-3-flash-preview" ,
contents : "Who won the 2024 Euro championship?" ,
config : {
tools : [{ googleSearch : {} }]
}
});
const { webSearchQueries, groundingChunks, groundingSupports } =
response.candidates [0 ].groundingMetadata ;
Extracting Citations (G5):
const citations = groundingChunks?.map (chunk =>url : chunk.web ?.uri ,
title : chunk.web ?.title
})) ?? [];
const supports = groundingSupports?.map (support =>text : support.segment ?.text ,
sources : support.groundingChunkIndices
})) ?? [];
config : {
tools : [
{ urlContext : {} },
{ googleSearch : {} }
]
}
Structured Output Force JSON responses matching a schema. G1: Always use responseJsonSchema — this guarantees valid JSON.
import { z } from "zod" ;
import { zodToJsonSchema } from "zod-to-json-schema" ;
const schema = z.object ({
company_name : z.string (),
industry : z.array (z.string ()),
description : z.string (),
founded_year : z.number ().optional ()
});
const response = await ai.models .generateContent ({
model : "gemini-3-flash-preview" ,
contents : `Extract company info from: ${url} ` ,
config : {
tools : [{ urlContext : {} }],
responseMimeType : "application/json" ,
responseJsonSchema : zodToJsonSchema (schema)
}
});
const data = schema.parse (JSON .parse (response.text ));
Why responseJsonSchema matters:
responseMimeType alone = model tries to output JSON (can fail)responseMimeType + responseJsonSchema = guaranteed syntactically valid JSON matching the schemaWorks with Google Search, URL Context, and Code Execution tools (G3)
Supported types: string, number, integer, boolean, object, array, null
REST API Patterns For raw fetch() calls in Edge Functions (instead of SDK):
Basic Request const GEMINI_API_KEY = Deno .env .get ("GEMINI_API_KEY" );
const model = "gemini-3-flash-preview" ;
const url = `https://generativelanguage.googleapis.com/v1beta/models/${model} :generateContent` ;
const response = await fetch (url, {
method : "POST" ,
headers : {
"Content-Type" : "application/json" ,
"x-goog-api-key" : GEMINI_API_KEY ,
},
body : JSON .stringify ({
contents : [{ parts : [{ text : prompt }] }],
generationConfig : {
responseMimeType : "application/json" ,
responseJsonSchema : schema,
thinkingConfig : {
thinkingLevel : "high"
}
}
})
});
const data = await response.json ();
const text = data.candidates ?.[0 ]?.content ?.parts ?.[0 ]?.text ;
With Google Search + Structured Output (G3) const response = await fetch (url, {
method : "POST" ,
headers : {
"Content-Type" : "application/json" ,
"x-goog-api-key" : GEMINI_API_KEY ,
},
body : JSON .stringify ({
contents : [{ parts : [{ text : prompt }] }],
tools : [{ googleSearch : {} }],
generationConfig : {
responseMimeType : "application/json" ,
responseJsonSchema : {
type : "object" ,
properties : {
summary : { type : "string" },
sources : {
type : "array" ,
items : {
type : "object" ,
properties : {
url : { type : "string" },
title : { type : "string" }
}
}
}
},
required : ["summary" , "sources" ]
}
}
})
});
const data = await response.json ();
const result = JSON .parse (data.candidates [0 ].content .parts [0 ].text );
const groundingChunks = data.candidates [0 ].groundingMetadata ?.groundingChunks ?? [];
Streaming (REST) const url = `https://generativelanguage.googleapis.com/v1beta/models/${model} :streamGenerateContent?alt=sse` ;
const response = await fetch (url, {
method : "POST" ,
headers : {
"Content-Type" : "application/json" ,
"x-goog-api-key" : GEMINI_API_KEY ,
},
body : JSON .stringify ({
contents : [{ parts : [{ text : prompt }] }]
})
});
Complete Edge Function Pattern import { GoogleGenAI } from "npm:@google/genai@^1.0.0" ;
import { z } from "npm:zod@^3.23.0" ;
import { zodToJsonSchema } from "npm:zod-to-json-schema@^3.23.0" ;
const ai = new GoogleGenAI ({ apiKey : Deno .env .get ("GEMINI_API_KEY" ) });
async function enrichUrl (url : string const schema = z.object ({
company_name : z.string (),
description : z.string (),
industry : z.array (z.string ()),
competitors : z.array (z.object ({
name : z.string (),
differentiator : z.string ()
}))
});
const response = await ai.models .generateContent ({
model : "gemini-3-flash-preview" ,
contents : `Analyze this company: ${url} . Find competitors.` ,
config : {
tools : [
{ urlContext : {} },
{ googleSearch : {} }
],
responseMimeType : "application/json" ,
responseJsonSchema : zodToJsonSchema (schema),
thinkingConfig : {
thinkingLevel : "medium"
}
}
});
const citations = response.candidates [0 ]?.groundingMetadata
?.groundingChunks ?.map (c =>url : c.web ?.uri , title : c.web ?.title })) ?? [];
return {
data : schema.parse (JSON .parse (response.text )),
citations
};
}
async function calculateReadiness (data : object const schema = z.object ({
overall_score : z.number ().min (0 ).max (100 ),
category_scores : z.object ({
product : z.number (),
market : z.number (),
team : z.number (),
clarity : z.number ()
}),
recommendations : z.array (z.string ())
});
const response = await ai.models .generateContent ({
model : "gemini-3-flash-preview" ,
contents : `Calculate investor readiness score: ${JSON .stringify(data)} ` ,
config : {
responseMimeType : "application/json" ,
responseJsonSchema : zodToJsonSchema (schema),
thinkingConfig : {
thinkingLevel : "high"
}
}
});
return schema.parse (JSON .parse (response.text ));
}
Function Calling Declare tools the model can call:
const tools = [{
functionDeclarations : [{
name : "get_weather" ,
description : "Get weather for a city" ,
parameters : {
type : "object" ,
properties : {
city : { type : "string" , description : "City name" }
},
required : ["city" ]
}
}]
}];
const response = await ai.models .generateContent ({
model : "gemini-3-flash-preview" ,
contents : "What's the weather in Paris?" ,
config : { tools }
});
const functionCall = response.functionCalls ?.[0 ];
if (functionCall) {
}
Modes: AUTO (default), ANY (force tool use), NONE (disable), VALIDATED
NOTE: Combining built-in tools (Google Search, URL Context) with function calling is NOT supported in Gemini 3.
Thought Signatures (Multi-turn) For function calling and image editing workflows, preserve thought context:
const response1 = await ai.models .generateContent ({...});
const signature = response1.candidates [0 ].content .parts [0 ].thoughtSignature ;
const response2 = await ai.models .generateContent ({
model : "gemini-3-flash-preview" ,
contents : [
{ role : "user" , parts : [{ text : "Initial prompt" }] },
{
role : "model" ,
parts : [{
functionCall : {...},
thoughtSignature : signature
}]
},
{ role : "user" , parts : [{ functionResponse : {...} }] }
]
});
Required for function calling (400 error without it)Required for multi-turn image editing conversations (single-turn generation works without them)Recommended for text/chat (improves reasoning continuity)SDKs handle this automatically
SDKs manage thought signatures automatically; preserve all accumulated signatures in multi-turn
Parallel calls: only the first functionCall has the signature
Multi-step: ALL accumulated signatures must be preserved
Image Generation const response = await ai.models .generateContent ({
model : "gemini-3-pro-image-preview" ,
contents : "Generate a logo for a tech startup" ,
config : {
tools : [{ googleSearch : {} }],
imageConfig : {
aspectRatio : "16:9" ,
imageSize : "4K"
}
}
});
const imagePart = response.candidates [0 ].content .parts
.find (p =>inlineData );
const imageData = imagePart.inlineData .data ;
4K resolution with text rendering
Grounded generation via Google Search
Conversational editing (multi-turn, requires thought signatures)
SynthID watermarks on all generated images
Document Processing Process PDFs and documents directly:
const response = await ai.models .generateContent ({
model : "gemini-3-flash-preview" ,
contents : [
{ text : "Summarize this document" },
{ inlineData : { mimeType : "application/pdf" , data : base64Data } }
]
});
const file = await ai.files .upload ({ file : "document.pdf" });
const response = await ai.models .generateContent ({
model : "gemini-3-flash-preview" ,
contents : [
{ text : "Extract key findings" },
{ fileData : { fileUri : file.uri , mimeType : file.mimeType } }
]
});
Supports up to 1000 pages
Use mediaResolution: "media_resolution_medium" for standard PDFs
Combine with structured output for typed extraction
System Instructions const response = await ai.models .generateContent ({
model : "gemini-3-flash-preview" ,
contents : "Hello there" ,
config : {
systemInstruction : "You are a startup advisor. Be concise and data-driven." ,
}
});
Temperature G2: Always use default temperature (1.0) for Gemini 3. Lower values cause looping or degraded performance, especially for reasoning and math tasks.
config : { temperature : 0.4 }
config : { }
Error Handling Code Error Cause Fix 400 INVALID_ARGUMENTBad request params, missing thought signature Check schema, ensure thought signatures present 403 PERMISSION_DENIEDInvalid API key or restricted Check key, ensure API enabled 404 NOT_FOUNDModel not found Verify model ID (e.g., gemini-3-flash-preview) 429 RESOURCE_EXHAUSTEDRate limit hit Implement exponential backoff 500 INTERNALServer error Retry with backoff 503 UNAVAILABLEService temporarily down Retry with backoff
async function callGeminiWithRetry (fn : () => Promise <any >, maxRetries = 3 for (let i = 0 ; i < maxRetries; i++) {
try {
return await fn ();
} catch (error) {
if (error.status === 429 || error.status >= 500 ) {
await new Promise (r =>setTimeout (r, Math .pow (2 , i) * 1000 ));
continue ;
}
throw error;
}
}
throw new Error ("Max retries exceeded" );
}
Token Counting const { thoughtsTokenCount, candidatesTokenCount, totalTokenCount } =
response.usageMetadata ;
const { toolUsePromptTokenCount } = response.usageMetadata ;
Common Patterns
Extraction with Fallback try {
const result = await enrichWithUrlContext (url);
} catch (error) {
const result = await enrichWithGoogleSearch (companyName);
}
Founder Lookup (Google Search Only)
const response = await ai.models .generateContent ({
model : "gemini-3-flash-preview" ,
contents : `Find info about founder: "${name} " at "${company} "` ,
config : {
tools : [{ googleSearch : {} }],
responseMimeType : "application/json" ,
responseJsonSchema : founderSchema
}
});
Resources
references/gemini-3.md - Full Gemini 3 documentationreferences/text-generation.md - Text gen, system instructions, temperaturereferences/structured-output.md - JSON schema output, Zod patternsreferences/google-search.md - Google Search grounding, citationsreferences/function-calling.md - Tool declarations, multi-turnreferences/thinking.md - Thinking levels, budgetsreferences/thought-signatures.md - Multi-turn state, function callingreferences/url-context.md - URL extraction, limitationsreferences/image-generation.md - Nano Banana, 4K, SynthIDreferences/document-processing.md - PDF, inline data, Files APIreferences/deep-search.md - Deep Research Agent, Interactions APIreferences/troubleshooting.md - Error codes, temperature warningsreferences/api-setup.md - API keys, SDKs, quickstart
Checklist When implementing Gemini AI:
