gemini-deep-research
Advanced deep research agent using Gemini 3.1 Pro with customtools endpoint for optimal agent performance. Deep research, thinking mode, and project management for 50+ SWAL projects. Updated 2026-04-20.
- License 许可
- MIT
- 所属仓库
- iberi22/skills
- 最近更新
- 2026-10-12
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- gemini-deep-research
- author
- swal
- license
- MIT
- updated
- 2026-04-23
- version
- 1.0.0
- metadata
- { "model": "gemini-3.1-pro-preview", "features": "thinking, code-execution, google-search, customtools", "provider": "Google AI Studio" }
- description
- Advanced deep research agent using Gemini 3.1 Pro with customtools endpoint for optimal agent performance. Deep research, thinking mode, and project management for 50+ SWAL projects. Updated 2026-04-20.
03 SKILL.md 正文
SWAL Deep Research Agent - Gemini 3.1 Pro Edition
🧠 Most Advanced Research System Using Gemini 3.1 Pro with customtools endpoint
🤖 Model Selection (UPDATED)
⭐ PRIMARY: gemini-3.1-pro-preview-customtools
- Released: February 19, 2026
- Why: Optimized for agent tool use - prioritizes custom tools over bash
- Endpoint: Native Google API (not OpenAI compatible)
- Cost: Same as standard
BACKUP: gemini-2.5-pro
- Why: Proven, stable
- Endpoint: OpenAI compatible available
🔑 API Keys (4 Keys Available)
Keys: AIzaSyBXo..., AIzaSyCxp..., AIzaSyDtx..., AIzaSyDj9...
🚨 IMPORTANT: API Endpoint
Gemini 3.1 uses NATIVE Google API, NOT OpenAI compatible:
// Native Google API (REQUIRED for 3.1)
POST https://generativelanguage.googleapis.com/v1beta/models/gemini-3.1-pro-preview:generateContent
// NOT the OpenAI endpoint!
POST https://generativelanguage.googleapis.com/v1beta/openai/chat/completions // ❌ Won't work
📡 API Configuration
Native Google API (for 3.1 Pro)
const response = await fetch(
`https://generativelanguage.googleapis.com/v1beta/models/gemini-3.1-pro-preview:generateContent?key=${API_KEY}`,
{
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({
contents: [{ parts: [{ text: prompt }] }],
generationConfig: {
maxOutputTokens: 8192,
temperature: 0.2
}
})
}
);
OpenAI Compatible (for 2.5 Flash)
// Works with OpenAI-compatible apps
POST https://generativelanguage.googleapis.com/v1beta/openai/chat/completions
🛠 CustomTools Endpoint
What it does:
- Better tool prioritization: Uses custom tools (view_file, search_code) instead of generic bash
- Less hallucinations: More reliable file operations
- Agent-optimized: Built for coding agents
Function Declarations (Recommended Names)
{
"tools": [{
"function_declarations": [
{
"name": "view_file",
"description": "Read a file from the filesystem",
"parameters": {
"type": "object",
"properties": {
"path": { "type": "string" }
}
}
},
{
"name": "search_code",
"description": "Search for code patterns",
"parameters": {
"type": "object",
"properties": {
"query": { "type": "string" }
}
}
},
{
"name": "execute_bash",
"description": "Execute a bash command",
"parameters": {
"type": "object",
"properties": {
"command": { "type": "string" }
}
}
}
]
}]
}
🔍 Available Models on Your Keys
| Model | Type | Available |
|---|---|---|
gemini-3.1-pro-preview |
Reasoning | ✅ |
gemini-3.1-pro-preview-customtools |
Agent Tools | ✅ |
gemini-3.1-flash-lite-preview |
Fast | ✅ |
gemini-2.5-pro |
Reasoning | ✅ |
gemini-2.5-flash |
Balance | ✅ |
🧠 Thinking Mode
Gemini 3.1 has enhanced thinking:
{
"model": "gemini-3.1-pro-preview",
"contents": {
"parts": [{ "text": "Your complex research query" }]
},
"generationConfig": {
"thinkingConfig": {
"thinkingBudgetTokens": 16384
}
}
}
🔄 Auto-Research Loop
┌─────────────────────────────────────────────────────────┐
│ DEEP RESEARCH LOOP │
├─────────────────────────────────────────────────────────┤
│ │
│ 1. Gemini 3.1 analyzes query │
│ 2. Generates search strategy │
│ 3. Uses google_search for web grounding │
│ 4. Synthesizes findings │
│ 5. Identifies gaps │
│ 6. Iterates if needed │
│ 7. Final comprehensive report │
│ │
└─────────────────────────────────────────────────────────┘
Research Prompt Template
const researchPrompt = `
PERFORM DEEP RESEARCH on: {TOPIC}
Requirements:
1. Use google_search for current information
2. Cite sources with URLs
3. Identify knowledge gaps
4. Provide actionable recommendations
Output: Comprehensive Markdown report
`;
📊 Rate Limit Strategy (4 Keys)
class ResearchRouter {
constructor() {
this.keys = [
'${GEMINI_API_KEY}',
'${GEMINI_API_KEY}',
'${GEMINI_API_KEY}',
'${GEMINI_API_KEY}'
];
this.index = 0;
}
getKey() {
const key = this.keys[this.index];
this.index = (this.index + 1) % this.keys.length;
return key;
}
}
🎯 Model Selection Guide
| Task | Model | Why |
|---|---|---|
| Coding Agent | 3.1-pro-customtools | Best tool use |
| Deep Research | 3.1-pro-preview | Best reasoning |
| Quick Analysis | 2.5-flash | Fast, cheap |
| Long Context | 3.1-pro (1M) | 1M tokens |
🚀 Quick Usage
Deep Research (Native API)
# Use the research script with 3.1
node scripts/deep-research-loop.js "AI agent frameworks 2026"
Coding Agent (CustomTools)
# Use OpenCode with customtools model
opencode run --model google/gemini-3.1-pro-preview-customtools "Create a React component"
Quick Analysis (OpenAI Compatible)
# Use Gemini CLI
gemini -p "Analyze this code architecture" -m gemini-2.5-flash
📁 Files
scripts/deep-research-loop.js- Auto-researchscripts/research-router.js- Key rotationscripts/swal-project-manager.js- Project tracking
Version: 2.0 (Gemini 3.1 Pro) Updated: 2026-04-20 Status: ✅ Ready with Gemini 3.1 Pro
04 同仓库更多 Skills
web-design-guidelines
暂无简介
blocked-page-recovery
Use when a fetch fails: 403/429, paywall, WAF, bot wall. Recovers the page via archives and provenance-checked fallbacks.
studio-coach
暂无简介
src-generator
暂无简介
sqlite-pro
暂无简介
rigor-evaluator
Use when you need a quick heuristic rigor score (10 criteria, 0-2 each) for the features in a .gitcore/features.json file, to decide which features are ready to ship.