zhongwei/gh-lucklyric-cc-dev-tools-plugins-gemini

Fork 0

Files

Zhongwei Li d2e5aed820 Initial commit

2025-11-30 08:38:21 +08:00

6.8 KiB

Raw Blame History

Gemini CLI Command Patterns

Purpose: Common command templates for Gemini CLI integration Version: v0.16.0+ Last Updated: 2025-11-18

Basic Invocation Patterns

One-Shot Queries (Headless Mode)

# Preferred syntax (positional prompt)
gemini -m gemini-3-pro-preview "Explain the observer pattern in software design"

# Alternative (deprecated -p flag, still works)
gemini -m gemini-3-pro-preview -p "Explain the observer pattern in software design"

# With stdin input
echo "function add(a, b) { return a + b; }" | gemini -m gemini-3-pro-preview "Explain this code"

Model Selection

# Use Gemini 3 Pro (default for complex reasoning)
gemini -m gemini-3-pro-preview "Design a microservices architecture"

# Use Gemini 2.5 Pro (stable, general reasoning)
gemini -m gemini-2.5-pro "Review this API design"

# Use Gemini 2.5 Flash (faster, code-focused)
gemini -m gemini-2.5-flash "Refactor this function for performance"

Version-Based Model Selection

When user requests a specific version, map to the latest model in that family:

# User says "use Gemini 3" or "use 3" → Latest 3.x Pro
gemini -m gemini-3-pro-preview "Design a distributed caching system"

# User says "use Gemini 2.5" for general tasks → 2.5 Pro
gemini -m gemini-2.5-pro "Explain the CAP theorem with examples"

# User says "use Gemini 2.5" for code editing → 2.5 Flash
gemini -m gemini-2.5-flash "Refactor this module for readability"

# No version specified → Default to latest Pro
gemini -m gemini-3-pro-preview "Research microservices best practices"

Fallback Strategy (when primary model unavailable):

# Try Gemini 3 Pro first
gemini -m gemini-3-pro-preview "Complex reasoning task" 2>&1

# If quota exhausted, fallback to 2.5 Pro (general) or 2.5 Flash (code)
if [ $? -ne 0 ]; then
    # For general reasoning
    gemini -m gemini-2.5-pro "Complex reasoning task"

    # OR for code editing
    # gemini -m gemini-2.5-flash "Complex reasoning task"
fi

Output Formatting

Text Output (Default)

gemini -m gemini-3-pro-preview "What is dependency injection?"

JSON Output (Programmatic)

gemini -m gemini-3-pro-preview --output-format json "List top 5 design patterns"

Streaming JSON (Real-time)

gemini -m gemini-3-pro-preview --output-format stream-json "Explain async/await in JavaScript"

Approval Modes & Security

Default Mode (Prompt for all actions)

# Explicit
gemini -m gemini-3-pro-preview --approval-mode default "Refactor this code"

# Implicit (no flag = default)
gemini -m gemini-3-pro-preview "Refactor this code"

Auto-Edit Mode (Auto-approve edit tools only)

gemini -m gemini-3-pro-preview --approval-mode auto_edit "Fix bugs in this file"

YOLO Mode (Auto-approve all tools)

# Long form
gemini -m gemini-3-pro-preview --approval-mode yolo "Deploy to production"

# Short form
gemini -m gemini-3-pro-preview -y "Deploy to production"

Sandbox Mode

Enable Sandbox

gemini -m gemini-3-pro-preview -s "Run untrusted code analysis"

Disable Sandbox (Default)

gemini -m gemini-3-pro-preview "Analyze trusted codebase"

Session Management

List Sessions

gemini --list-sessions

Resume Sessions

# Resume most recent session
gemini -r latest

# Resume specific session by index
gemini -r 3

# Resume and add new prompt
gemini -r latest "Continue from where we left off"

Delete Sessions

gemini --delete-session 5

Extensions & MCP

List Available Extensions

gemini -l
# or
gemini --list-extensions

Use Specific Extensions

gemini -e web_search,code_analysis "Research React best practices"

Use All Extensions (Default)

gemini "Research React best practices"

Workspace Context

Include Additional Directories

gemini --include-directories ./lib,./tests "Analyze the full project"

# Or multiple flags
gemini --include-directories ./lib --include-directories ./tests "Analyze project"

Combined Patterns

Production-Safe Code Review

gemini -m gemini-2.5-pro \
  --approval-mode default \
  --output-format json \
  "Review this pull request for security issues"

Fast Code Refactoring

gemini -m gemini-2.5-flash \
  --approval-mode auto_edit \
  "Refactor these functions for better performance"

Research with Web Search

gemini -m gemini-3-pro-preview \
  -e web_search \
  --output-format text \
  "What are the latest trends in GraphQL?"

Sandbox Testing

gemini -m gemini-2.5-pro \
  -s \
  --approval-mode default \
  "Test this suspicious code snippet"

Claude Code Integration Patterns

Skill Invocation (via Claude)

# Basic invocation from skill
gemini -m gemini-3-pro-preview "Explain microservices"

# With explicit approval mode
gemini -m gemini-2.5-pro --approval-mode auto_edit "Fix type errors"

# With JSON output for parsing
gemini -m gemini-2.5-flash --output-format json "List API endpoints"

Error Handling Pattern

# Check CLI availability
if ! command -v gemini &> /dev/null; then
    echo "Error: Gemini CLI not installed"
    exit 1
fi

# Execute with error capture
gemini -m gemini-3-pro-preview "task" 2>&1 || echo "Gemini CLI failed"

Model Selection by Task Type

Complex Reasoning (General)

gemini -m gemini-3-pro-preview "Design system architecture"

Code Review & Analysis

gemini -m gemini-2.5-pro "Review this pull request"

Fast Code Editing

gemini -m gemini-2.5-flash "Fix syntax errors in this file"

Anti-Patterns (Avoid These)

# ❌ Using -p flag (deprecated)
gemini -p "prompt"  # Will be removed in future

# ❌ Using -i for headless mode
gemini -i "prompt"  # This starts interactive mode, not one-shot

# ❌ Hardcoding model without fallback
# Always have fallback logic when gemini-3-pro-preview unavailable

# ❌ Using YOLO mode without user confirmation
# Always require explicit user approval for YOLO mode

Best Practices

Prefer Positional Prompts: Use gemini "prompt" instead of gemini -p "prompt"
Specify Model Explicitly: Always use -m flag for predictable behavior
Use Appropriate Approval Mode: Default for untrusted tasks, auto_edit for code editing
Enable Sandbox for Unknown Code: Use -s when analyzing untrusted input
Format Output for Parsing: Use --output-format json when processing results programmatically
Resume Sessions When Needed: Use -r latest for multi-turn conversations
Validate CLI Availability: Always check command -v gemini before invocation

6.8 KiB Raw Blame History