gemini-search

tt44mm's avatarfrom tt44mm

Performs automatic web search using Gemini CLI with Google Search grounding. Triggers on "Gemini search:", "suche mit Gemini", "use Gemini to find". Claude writes a request file, the Watcher executes the search, Claude reads the result. FREE Gemini 2.5 Pro with 1000 requests/day.

⭐0stars🔀0forks📁View on GitHub🕐Updated Jan 2, 2026
[Web Research]

When & Why to Use This Skill

This Claude skill enables fully automated web research by integrating the Gemini CLI with Google Search grounding. It bridges the gap between Claude and real-time internet data, allowing the agent to write search requests, trigger a local watcher, and ingest high-quality search results. Leveraging Gemini 2.5 Pro with a generous free tier of 1000 requests per day, it provides a robust solution for fetching up-to-date information directly into the Claude environment.

Use Cases

  • Real-time Fact Checking: Verify current events, news, or statistics by performing live searches with Google Search grounding to ensure response accuracy.
  • Market & Trend Analysis: Automate the gathering of the latest industry reports, competitor news, and market trends to synthesize up-to-date business intelligence.
  • Technical Research: Retrieve the latest documentation, API changes, or library updates that occurred after Claude's last training cutoff.
  • Automated Report Generation: Streamline workflows where Claude needs to search for multiple data points across the web to compile comprehensive research summaries.
namegemini-search
descriptionPerforms automatic web search using Gemini CLI with Google Search grounding. Triggers on "Gemini search:", "suche mit Gemini", "use Gemini to find". Claude writes a request file, the Watcher executes the search, Claude reads the result. FREE Gemini 2.5 Pro with 1000 requests/day.

Gemini Search Skill (Automatischer Workflow)

Vollautomatische Web-Recherche mit Gemini CLI und Google Search Grounding.

Workflow

Schritt 1: Request-Datei schreiben

Claude schreibt eine JSON-Datei nach:

X:\DEVELOPMENTS\gemini_search\requests\search.request.json

Inhalt:

{
  "id": "unique-id",
  "query": "Die Suchanfrage",
  "format": "json"
}

Format-Optionen: "json" oder "md"

Schritt 2: Watcher verarbeitet automatisch

Der Watcher (muss im Hintergrund laufen!) erkennt die Datei und:

  1. Führt Gemini CLI aus
  2. Speichert Ergebnis in results/gemini_result.json
  3. Schreibt results/ready.flag als Signal

Schritt 3: Auf Ergebnis warten

Claude prüft X:\DEVELOPMENTS\gemini_search\results\status.json:

  • "status": "idle" → Watcher bereit, kein Request
  • "status": "searching" → Suche läuft, warten
  • "status": "complete" → Ergebnis bereit!
  • "status": "error" → Fehler aufgetreten

Schritt 4: Ergebnis lesen

Wenn status = complete, lies:

X:\DEVELOPMENTS\gemini_search\results\gemini_result.json

Wichtige Pfade

X:\DEVELOPMENTS\gemini_search\
├── requests\                    ← Claude schreibt hier
│   └── search.request.json
├── results\                     ← Claude liest hier
│   ├── gemini_result.json       ← Suchergebnis
│   ├── gemini_result.md         ← Markdown (falls format=md)
│   ├── status.json              ← Aktueller Status
│   └── ready.flag               ← Signal: Ergebnis bereit
├── gemini_watcher.py            ← Hintergrund-Service
├── start_watcher.bat            ← Watcher starten (sichtbar)
├── start_watcher_hidden.bat     ← Watcher starten (unsichtbar)
└── watcher.log                  ← Log-Datei

Voraussetzungen

  1. Watcher muss laufen! Starte einmalig:

    X:\DEVELOPMENTS\gemini_search\start_watcher.bat

  2. Gemini CLI installiert und eingeloggt

Beispiel Request

{
  "id": "search-20250102-143000",
  "query": "Aktuelle Entwicklungen KI in der Schulbildung Deutschland 2025",
  "format": "json"
}

Status-Datei Format

{
  "status": "complete",
  "query": "...",
  "completed": "2025-01-02T14:30:45",
  "request_id": "search-20250102-143000",
  "result_file": "X:\\...\\gemini_result.json"
}

Timeout / Polling

  • Claude sollte max 60 Sekunden auf Ergebnis warten
  • Prüfe status.json alle 3-5 Sekunden
  • Bei Timeout: User informieren, Watcher-Status prüfen