Claude Code Plugins

Community-maintained marketplace

Feedback

Guide for building Gemini CLI TOML custom commands. Covers syntax, templates, argument handling, shell injection, and file injection. Use when creating Gemini TOML commands, adding {{args}} argument handling, injecting shell output with !{}, or troubleshooting command issues.

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 toml-command-builder
description Guide for building Gemini CLI TOML custom commands. Covers syntax, templates, argument handling, shell injection, and file injection. Use when creating Gemini TOML commands, adding {{args}} argument handling, injecting shell output with !{}, or troubleshooting command issues.
allowed-tools Read, Glob, Grep, Skill

TOML Command Builder

🚨 MANDATORY: Invoke gemini-cli-docs First

STOP - Before providing ANY response about TOML commands:

  1. INVOKE gemini-cli-docs skill
  2. QUERY for the specific command topic
  3. BASE all responses EXCLUSIVELY on official documentation loaded

Overview

This skill provides comprehensive guidance for creating Gemini CLI custom commands using TOML format. Custom commands are slash commands that extend Gemini's capabilities with project-specific or user-specific functionality.

When to Use This Skill

Keywords: toml command, custom command, slash command gemini, gemini command, create command, {{args}}, @{}, !{}

Use this skill when:

  • Creating a new Gemini CLI custom command
  • Understanding TOML command syntax
  • Adding argument handling to commands
  • Injecting shell output or file content
  • Troubleshooting command issues

Command Locations

User Commands (Global)

~/.gemini/commands/
├── commit.toml           # /commit
├── review.toml           # /review
└── git/
    └── log.toml          # /git:log (namespaced)

Project Commands (Local)

.gemini/commands/
├── build.toml            # /build
├── test.toml             # /test
└── deploy/
    └── staging.toml      # /deploy:staging

Basic Syntax

Minimal Command

# ~/.gemini/commands/hello.toml
description = "A simple greeting command"
prompt = "Say hello to the user"

With Multi-line Prompt

description = "Generate a commit message"
prompt = """
Analyze the staged changes and generate a commit message.

Follow conventional commit format:
- feat: new features
- fix: bug fixes
- docs: documentation
- refactor: code refactoring
"""

Argument Handling

Basic Arguments ({{args}})

Arguments passed after the command are injected at {{args}}:

# /greet Kyle -> "Say hello to Kyle"
description = "Greet a person"
prompt = "Say hello to {{args}}"

Default Behavior

If no {{args}} placeholder, arguments are appended:

description = "Analyze code"
prompt = "Analyze this code for issues"
# /analyze src/main.ts -> "Analyze this code for issues src/main.ts"

Multiple Arguments

Arguments are space-separated, accessible together:

description = "Compare two files"
prompt = "Compare these files: {{args}}"
# /compare file1.ts file2.ts -> "Compare these files: file1.ts file2.ts"

Shell Injection (!{...})

Execute shell commands and inject output:

Basic Shell

description = "Analyze git diff"
prompt = """
Review the following git diff:

~~~diff
!{git diff --staged}
~~~

Suggest improvements.
"""

Shell with Safety

Commands are confirmed before execution. Use --yolo to skip.

Complex Shell

description = "Analyze project structure"
prompt = """
Project structure:
!{find . -type f -name "*.ts" | head -50}

Package dependencies:
!{cat package.json | jq '.dependencies'}

Analyze and suggest improvements.
"""

Shell with Arguments

Combine shell injection with arguments:

description = "Grep for pattern"
prompt = """
Search results for "{{args}}":

!{grep -r "{{args}}" src/ --include="*.ts" | head -20}

Analyze these occurrences.
"""

Note: Arguments in shell blocks are automatically escaped for safety.

File Injection (@{...})

Inject file or directory contents:

Single File

description = "Review config"
prompt = """
Review this configuration:

@{tsconfig.json}

Suggest improvements.
"""

Multiple Files

description = "Review setup"
prompt = """
Package config:
@{package.json}

TypeScript config:
@{tsconfig.json}

Analyze for consistency.
"""

Directory Contents

description = "Review utilities"
prompt = """
Utility functions:
@{src/utils/}

Analyze for patterns and improvements.
"""

Note: Directory injection respects .gitignore and .geminiignore.

With Arguments

description = "Review file"
prompt = """
Review this file:
@{{{args}}}

Provide feedback.
"""
# /review src/main.ts -> Injects content of src/main.ts

Processing Order

Injection is processed in order:

  1. @{...} - File content injection
  2. !{...} - Shell command execution
  3. {{args}} - Argument substitution

Namespacing

Organize commands with directories:

~/.gemini/commands/
├── git/
│   ├── commit.toml     # /git:commit
│   ├── review.toml     # /git:review
│   └── log.toml        # /git:log
├── test/
│   ├── unit.toml       # /test:unit
│   └── e2e.toml        # /test:e2e
└── deploy/
    ├── staging.toml    # /deploy:staging
    └── prod.toml       # /deploy:prod

Template Library

Git Commit Message

# ~/.gemini/commands/git/commit.toml
description = "Generate conventional commit message from staged changes"
prompt = """
Analyze the staged changes and generate a commit message.

## Staged Changes
~~~diff
!{git diff --staged}
~~~

## Requirements

- Use conventional commit format (feat/fix/docs/refactor/test/chore)
- Keep subject line under 72 characters
- Add body if changes are significant
- Reference issue numbers if applicable

Generate only the commit message, nothing else.
"""

Code Review

# ~/.gemini/commands/review.toml
description = "Review code changes with specific focus"
prompt = """
Review the following code changes:

~~~diff
!{git diff}
~~~

Focus on: {{args}}

Provide:

1. Issues found (if any)
2. Suggestions for improvement
3. Positive observations
"""

Test Generator

# ~/.gemini/commands/test/generate.toml
description = "Generate tests for a file"
prompt = """
Generate comprehensive tests for this file:

@{{{args}}}

Requirements:
- Use the existing test framework (jest/vitest/pytest)
- Cover edge cases
- Include positive and negative tests
- Follow existing test patterns in the project
"""

Documentation Generator

# ~/.gemini/commands/docs/generate.toml
description = "Generate documentation for code"
prompt = """
Generate documentation for:

@{{{args}}}

Include:
- Purpose and overview
- Function/method documentation
- Usage examples
- Parameter descriptions
"""

Dependency Analyzer

# ~/.gemini/commands/deps/analyze.toml
description = "Analyze project dependencies"
prompt = """
Analyze these dependencies:

## package.json
@{package.json}

## Lock file (partial)
!{head -100 package-lock.json 2>/dev/null || head -100 yarn.lock 2>/dev/null || echo "No lock file"}

Identify:
1. Outdated packages
2. Security concerns
3. Unused dependencies
4. Duplicate functionality
"""

Migration Helper

# ~/.gemini/commands/migrate.toml
description = "Help with code migration"
prompt = """
Help migrate this code: {{args}}

Current code:
@{{{args}}}

Migration requirements:
- Preserve functionality
- Follow modern patterns
- Add TypeScript types if missing
- Update deprecated APIs
"""

Validation Patterns

Check Syntax

# Validate TOML syntax
python -c "import tomllib; tomllib.load(open('command.toml', 'rb'))"

Required Fields

  • description (string): Shown in command list
  • prompt (string): The prompt template

Common Errors

Error Cause Fix
Parse error Invalid TOML syntax Check quotes, brackets
Command not found Wrong location Verify path
Args not injected Missing {{args}} Add placeholder
Shell fails Command error Test command manually

Best Practices

1. Clear Descriptions

# Good
description = "Generate commit message from staged changes using conventional format"

# Bad
description = "commit stuff"

2. Structured Prompts

prompt = """
## Task
{what to do}

## Context
{relevant information}

## Requirements
- {requirement 1}
- {requirement 2}

## Output Format
{expected format}
"""

3. Safe Shell Commands

# Good - read-only, limited output
!{git diff --staged | head -500}

# Risky - potentially destructive
!{rm -rf {{args}}}  # DANGEROUS!

4. Helpful Error Messages

prompt = """
{{args}}

If no file specified, respond: "Please specify a file path after the command"
"""

5. Consistent Naming

commands/
├── git/          # Git operations
│   ├── commit.toml
│   └── review.toml
├── test/         # Testing
│   ├── unit.toml
│   └── e2e.toml
└── docs/         # Documentation
    └── generate.toml

Related Skills

  • gemini-cli-docs - Official command documentation
  • policy-engine-builder - Tool execution policies

Related Commands

  • /build-toml-command - Interactive command builder wizard

Keyword Registry

Topic Keywords
Basic syntax toml command, custom command, gemini command
Arguments {{args}}, command arguments, argument injection
Shell !{...}, shell injection, git diff command
Files @{...}, file injection, directory contents
Namespacing command namespace, git:commit, organize commands

Test Scenarios

Scenario 1: Create Basic Command

Query: "How do I create a custom TOML command in Gemini CLI?" Expected Behavior:

  • Skill activates on "toml command" or "custom command"
  • Provides basic TOML syntax with description and prompt Success Criteria: User receives working .toml file template

Scenario 2: Shell Injection

Query: "How do I include git diff in a Gemini command?" Expected Behavior:

  • Skill activates on "shell injection" or "git diff"
  • Provides !{git diff} syntax Success Criteria: User receives shell injection pattern with safety notes

Scenario 3: File Injection

Query: "How do I inject file contents into a Gemini command?" Expected Behavior:

  • Skill activates on "file injection" or "@{"
  • Provides @{filename} syntax Success Criteria: User receives file injection pattern with directory support

Version History

  • v1.1.0 (2025-12-01): Added MANDATORY section, Test Scenarios, Version History
  • v1.0.0 (2025-11-25): Initial release