Claude Code Plugins

Community-maintained marketplace

Feedback

gemini-search

@22GNUs/dotfiles
1
0

Use Gemini CLI for Google Search and deep web research. Triggers on phrases like "search", "look up", "research", "latest info", "find out", "google", or when current/real-time information is needed.

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name gemini-search
description Use Gemini CLI for Google Search and deep web research. Triggers on phrases like "search", "look up", "research", "latest info", "find out", "google", or when current/real-time information is needed.
license MIT
compatibility opencode

Gemini Search Skill

Leverage Gemini CLI to perform Google Search with optimized modes for different use cases.

Capabilities

  1. Quick Search - Uses gemini-2.5-flash-lite for fast Google Search queries
  2. Deep Research - Uses gemini-3-flash-preview for comprehensive analysis with extended thinking

Both scripts support parallel execution for multiple concurrent searches.

Usage

Quick Search (for simple queries)

Execute with Bash tool:

~/.config/opencode/skill/gemini-search/scripts/quick_search.sh "your search query"

Best for:

  • Looking up API documentation
  • Finding quick factual information
  • Searching for error message solutions
  • Verifying version numbers or dates

Deep Research (for complex topics)

Execute with Bash tool:

~/.config/opencode/skill/gemini-search/scripts/deep_research.sh "your research question"

Best for:

  • Technical architecture research
  • Multi-perspective problem analysis
  • Topics requiring synthesis from multiple sources
  • Comprehensive technology comparisons

Parallel Execution

Both scripts can be run concurrently for multiple queries:

# Run multiple quick searches in parallel
quick_search.sh "query1" &
quick_search.sh "query2" &
quick_search.sh "query3" &
wait

# Mix quick and deep searches
quick_search.sh "what is rust" &
deep_research.sh "rust vs go performance comparison" &
wait

When running multiple searches in parallel, all queries execute simultaneously and complete in the time of the slowest query.

Output Format

Results include structured information with confidence indicators:

Quick Search Output:

ANSWER: [Direct answer]

FACTS:
- [statement] [FACT|CONFIDENCE] (source: [url])
- [statement] [OPINION|CONFIDENCE] (source: [url])

Deep Research Output:

SUMMARY: [Overview]

KEY_FINDINGS:
- [statement] [FACT|CONFIDENCE] (source: [url])
- [statement] [OPINION|CONFIDENCE] (source: [url])

ANALYSIS: [Deeper insights]

SOURCES:
- [url]: [description]

Confidence Levels:

  • HIGH: Verified by multiple authoritative sources
  • MEDIUM: Single authoritative source or multiple less authoritative
  • LOW: Unverified or conflicting sources

Fact vs Opinion:

  • FACT: Objectively verifiable (dates, numbers, definitions, technical specs)
  • OPINION: Subjective assessments, interpretations, recommendations

Performance

  • Quick search: ~10-30s (optimized for speed)
  • Deep research: ~30-60s (optimized for comprehensiveness)
  • Parallel execution: Same as single search (runs concurrently)

Performance optimizations:

  • Uses gemini-2.5-flash-lite for quick searches (fastest model)
  • Headless mode with JSON output for minimal overhead
  • Compact prompts to reduce token usage
  • Suppressed debug logs for clean output

Requirements

  • Gemini CLI installed (npm install -g @anthropic-ai/gemini-cli)
  • Logged into Gemini CLI (run gemini for first-time login)

Notes

  • Scripts already exist, do not create them
  • All searches use headless mode for clean, parseable output
  • Results are optimized for AI agent consumption
  • Deep research mode uses extended thinking for better analysis
  • Both scripts are stateless and safe for concurrent execution