### Agent TOML Configuration Schema Example

Source: https://context7.com/jduncan-rva/gemini-agent-creator/llms.txt

An example of a basic agent definition file in TOML format, illustrating required fields like name, displayName, description, icon, tools, systemPrompt, query, and inputs.

```toml
# ~/.gemini/agents/test_writer.toml

name = "test_writer"                          # snake_case, unique, used as tool name
displayName = "🧪 Test Writer"               # Human-friendly name shown in CLI
description = "Generates unit tests for code" # One sentence
icon = "🧪"                                   # Single emoji

```

--------------------------------

### Install Agent Creator Extension

Source: https://context7.com/jduncan-rva/gemini-agent-creator/llms.txt

Clone the extension repository into the Gemini CLI extensions directory and restart the CLI. Verify the installation using the '/extensions list' command.

```bash
# Install the extension
git clone https://github.com/jduncan-rva/gemini-agent-creator.git ~/.gemini/extensions/agent-creator

# Restart Gemini CLI, then verify the extension loaded
/extensions list
# Expected output includes: agent-creator
```

--------------------------------

### Agent Configuration Example

Source: https://context7.com/jduncan-rva/gemini-agent-creator/llms.txt

Configure agent model, temperature, max time, max turns, allowed tools, system prompt, and input/output parameters. Ensure tools are from the allowed list.

```toml
model = "gemini-2.5-pro"                      # or "gemini-2.0-flash-thinking-exp"
temperature = 0.1                             # 0.0–2.0; higher = more creative
maxTimeMinutes = 5
maxTurns = 15

# Allowed tools (only these values are valid)
# ls | read | grep | glob | read_many_files | memory | web_search
tools = ["read", "grep", "glob"]

systemPrompt = """
You are a test generation expert.

Generate comprehensive unit tests covering:
- Happy path scenarios
- Edge cases and boundary conditions
- Error handling and exceptions
- Mocking of external dependencies

Follow the existing test framework conventions found in the project.
"""

query = "Generate tests for ${target_file} using ${test_framework}"

[inputs.target_file]
type = "string"
description = "Path to the file to generate tests for"
required = true

[inputs.test_framework]
type = "string"
description = "Test framework to use (e.g., pytest, jest, go test)"
required = false

[output]
name = "result"
description = "Generated test file content"
type = "string"
```

--------------------------------

### Install Agent Creator Extension

Source: https://github.com/jduncan-rva/gemini-agent-creator/blob/main/README.md

Clone the repository into your Gemini extensions directory and restart the CLI to load the extension.

```bash
git clone https://github.com/jduncan-rva/gemini-agent-creator.git ~/.gemini/extensions/agent-creator
```

--------------------------------

### Verify Agent Creator Installation

Source: https://github.com/jduncan-rva/gemini-agent-creator/blob/main/README.md

Run this command to check if the agent-creator extension is loaded in your Gemini CLI.

```bash
/extensions list
```

--------------------------------

### Create Agent with Interactive Mode

Source: https://github.com/jduncan-rva/gemini-agent-creator/blob/main/README.md

Run the command without arguments to be guided step-by-step through the agent creation process.

```bash
/new-agent
```

--------------------------------

### Example Agent TOML Configuration

Source: https://github.com/jduncan-rva/gemini-agent-creator/blob/main/README.md

This TOML configuration defines an agent for performance analysis, including its name, display name, icon, tools, system prompt, and query.

```toml
name = "performance_analyzer"
displayName = "⚡ Performance Analyzer"
description = "Analyzes code for performance issues and suggests optimizations"
icon = "⚡"

tools = ["read", "grep", "glob"]

systemPrompt = """
You are a performance optimization expert...
[detailed instructions]"""

query = "Analyze ${file_path} for performance issues"

[inputs.file_path]
type = "string"
description = "Path to file or directory to analyze"
required = true
```

--------------------------------

### Invoking Created Agents via CLI

Source: https://context7.com/jduncan-rva/gemini-agent-creator/llms.txt

Examples of how to invoke agents using their names and providing necessary arguments. Use '/agents' to list available agents and '$EDITOR' to edit agent configurations.

```bash
# Restart CLI to load newly created agents
# Then invoke the agent by its name:

/security_reviewer file_path=src/auth/login.py
# → AI reviews the file for SQL injection, XSS, auth issues, etc.

/api_doc_generator target_path=src/api/
# → AI generates markdown documentation for all files in src/api/

/test_writer target_file=src/utils/parser.py test_framework=pytest
# → AI generates a pytest test file for parser.py

# List all available agents:
/agents

# Edit an agent manually:
# $EDITOR ~/.gemini/agents/security_reviewer.toml

# Delete and recreate an agent:
# rm ~/.gemini/agents/security_reviewer.toml
# /new-agent "reviews code for security vulnerabilities"
```

--------------------------------

### Create Agent - Interactive Mode

Source: https://context7.com/jduncan-rva/gemini-agent-creator/llms.txt

Run '/new-agent' without arguments to enter an interactive mode where the AI guides you step-by-step through defining each agent field. The AI suggests defaults and validates entries before generating the final TOML.

```bash
/new-agent

# AI: What kind of agent would you like to create?
# User: I want something that generates API documentation from source code

# AI: Great! Here are my suggestions:
#   name: "api_doc_generator"
#   displayName: "📚 API Doc Generator"
#   tools: ["read", "read_many_files", "glob"]

# After completing all steps, the AI generates:

name = "api_doc_generator"
displayName = "📚 API Doc Generator"
description = "Generates comprehensive API documentation from source code"
icon = "📚"

model = "gemini-2.5-pro"
temperature = 0.1
maxTimeMinutes = 5
maxTurns = 15

tools = ["read", "read_many_files", "glob"]

systemPrompt = """
You are a technical documentation expert.

Generate clear, comprehensive documentation including:
- Function/class descriptions
- Parameter documentation
- Return value documentation
- Usage examples
- Edge cases and limitations

Format as markdown with proper headings and code blocks.
"""

query = "Generate documentation for ${target_path}"

[inputs.target_path]
type = "string"
description = "Path to file or directory to document"
required = true

[inputs.format]
type = "string"
description = "Output format (default: markdown)"
required = false

[output]
name = "result"
description = "The generated documentation"
type = "string"

# Saved to: ~/.gemini/agents/api_doc_generator.toml
```

--------------------------------

### Create Agent - Quick Mode

Source: https://context7.com/jduncan-rva/gemini-agent-creator/llms.txt

Use the '/new-agent' command with a one-line description to have the AI infer and generate a complete agent TOML configuration. The AI suggests name, icon, tools, system prompt, inputs, and query, then prompts for approval before saving.

```bash
/new-agent "reviews code for security vulnerabilities"
```

```toml
# AI responds with the proposed configuration:

name = "security_reviewer"
displayName = "🔒 Security Reviewer"
description = "Reviews code for security vulnerabilities and suggests fixes"
icon = "🔒"

tools = ["read", "grep", "glob"]

systemPrompt = """
You are a security expert specializing in code review.

Analyze code for:
- SQL injection vulnerabilities
- XSS (Cross-Site Scripting) risks
- Authentication/authorization issues
- Insecure data handling
- Dependency vulnerabilities
- Sensitive data exposure

For each issue:
1. Explain the vulnerability
2. Show the problematic code
3. Provide a secure alternative
4. Rate severity (Critical/High/Medium/Low)
"""

query = "Review ${file_path} for security vulnerabilities"

[inputs.file_path]
type = "string"
description = "Path to file or directory to review"
required = true

# After user approval, the file is saved to:
# ~/.gemini/agents/security_reviewer.toml
```

--------------------------------

### Create Agent with Quick Mode

Source: https://github.com/jduncan-rva/gemini-agent-creator/blob/main/README.md

Provide a brief description of the desired agent functionality to have the AI generate a complete TOML configuration.

```bash
/new-agent "reviews code for security vulnerabilities"
```

--------------------------------

### Available Agent Tools

Source: https://context7.com/jduncan-rva/gemini-agent-creator/llms.txt

List of all available read-only tools for agents. Select only the tools your agent requires. Using unlisted tools will result in validation failure.

```toml
# All available tools — select only what your agent actually needs

tools = [
  "ls",              # List directory contents
  "read",            # Read a single file's contents
  "grep",            # Search for text patterns within files
  "glob",            # Find files matching a pattern (e.g., "**/*.py")
  "read_many_files", # Read multiple files in a single operation
  "memory",          # Access and store persistent memory across sessions
  "web_search"       # Search the web for current information
]

# Recommended tool sets by agent type:
# Code analysis:   ["read", "grep", "glob"]
# Documentation:   ["read", "read_many_files", "web_search"]
# Research:        ["read", "grep", "web_search", "memory"]
# File discovery:  ["ls", "glob", "read_many_files"]
```

--------------------------------

### Extension Manifest File

Source: https://context7.com/jduncan-rva/gemini-agent-creator/llms.txt

The `gemini-extension.json` file registers the extension with the Gemini CLI. Ensure the 'name' field matches the extension directory name and restart the CLI after editing.

```json
{
  "name": "agent-creator",
  "version": "1.0.0"
}

// Located at: ~/.gemini/extensions/agent-creator/gemini-extension.json
// The "name" field must match the extension directory name.
// After editing this file, restart Gemini CLI to apply changes.
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.