Skill Creator

This skill provides guidance for creating effective skills.

About Skills

Skills are modular, self-contained packages that extend Gemini CLI’s capabilities by providing specialized knowledge, workflows, and tools. Think of them as “onboarding guides” for specific domains or tasks—they transform Gemini CLI from a general-purpose agent into a specialized agent equipped with procedural knowledge that no model can fully possess.

What Skills Provide

1. Specialized workflows - Multi-step procedures for specific domains
2. Tool integrations - Instructions for working with specific file formats or APIs
3. Domain expertise - Company-specific knowledge, schemas, business logic
4. Bundled resources - Scripts, references, and assets for complex and repetitive tasks

Core Principles

Concise is Key

The context window is a public good. Skills share the context window with everything else Gemini CLI needs: system prompt, conversation history, other Skills’ metadata, and the actual user request.

Default assumption: Gemini CLI is already very smart. Only add context Gemini CLI doesn’t already have. Challenge each piece of information: “Does Gemini CLI really need this explanation?” and “Does this paragraph justify its token cost?”

Prefer concise examples over verbose explanations.

Set Appropriate Degrees of Freedom

Match the level of specificity to the task’s fragility and variability:

High freedom (text-based instructions): Use when multiple approaches are valid, decisions depend on context, or heuristics guide the approach.

Medium freedom (pseudocode or scripts with parameters): Use when a preferred pattern exists, some variation is acceptable, or configuration affects behavior.

Low freedom (specific scripts, few parameters): Use when operations are fragile and error-prone, consistency is critical, or a specific sequence must be followed.

Think of Gemini CLI as exploring a path: a narrow bridge with cliffs needs specific guardrails (low freedom), while an open field allows many routes (high freedom).

Anatomy of a Skill

Every skill consists of a required SKILL.md file and optional bundled resources:

skill-name/
├── SKILL.md (required)
│   ├── YAML frontmatter metadata (required)
│   │   ├── name: (required)
│   │   └── description: (required)
│   └── Markdown instructions (required)
└── Bundled Resources (optional)
    ├── scripts/          - Executable code (Node.js/Python/Bash/etc.)
    ├── references/       - Documentation intended to be loaded into context as needed
    └── assets/           - Files used in output (templates, icons, fonts, etc.)

SKILL.md (required)

Every SKILL.md consists of:

• Frontmatter (YAML): Contains name and description fields. These are the only fields that Gemini CLI reads to determine when the skill gets used, thus it is very important to be clear and comprehensive in describing what the skill is, and when it should be used.
• Body (Markdown): Instructions and guidance for using the skill. Only loaded AFTER the skill triggers (if at all).

Bundled Resources (optional)

Scripts (scripts/)

Executable code (Node.js/Python/Bash/etc.) for tasks that require deterministic reliability or are repeatedly rewritten.

• When to include: When the same code is being rewritten repeatedly or deterministic reliability is needed
• Example: scripts/rotate_pdf.cjs for PDF rotation tasks
• Benefits: Token efficient, deterministic, may be executed without loading into context
• Agentic Ergonomics: Scripts must output LLM-friendly stdout. Suppress standard tracebacks. Output clear, concise success/failure messages, and paginate or truncate outputs (e.g., “Success: First 50 lines of processed file…”) to prevent context window overflow.
• Note: Scripts may still need to be read by Gemini CLI for patching or environment-specific adjustments

References (references/)

Documentation and reference material intended to be loaded as needed into context to inform Gemini CLI’s process and thinking.

• When to include: For documentation that Gemini CLI should reference while working
• Examples: references/finance.md for financial schemas, references/mnda.md for company NDA template, references/policies.md for company policies, references/api_docs.md for API specifications
• Use cases: Database schemas, API documentation, domain knowledge, company policies, detailed workflow guides
• Benefits: Keeps SKILL.md lean, loaded only when Gemini CLI determines it’s needed
• Best practice: If files are large (>10k words), include grep search patterns in SKILL.md
• Avoid duplication: Information should live in either SKILL.md or references files, not both. Prefer references files for detailed information unless it’s truly core to the skill—this keeps SKILL.md lean while making information discoverable without hogging the context window. Keep only essential procedural instructions and workflow guidance in SKILL.md; move detailed reference material, schemas, and examples to references files.

Assets (assets/)

Files not intended to be loaded into context, but rather used within the output Gemini CLI produces.

• When to include: When the skill needs files that will be used in the final output
• Examples: assets/logo.png for brand assets, assets/slides.pptx for PowerPoint templates, assets/frontend-template/ for HTML/React boilerplate, assets/font.ttf for typography
• Use cases: Templates, images, icons, boilerplate code, fonts, sample documents that get copied or modified
• Benefits: Separates output resources from documentation, enables Gemini CLI to use files without loading them into context

What to Not Include in a Skill

A skill should only contain essential files that directly support its functionality. Do NOT create extraneous documentation or auxiliary files, including:

• README.md
• INSTALLATION_GUIDE.md
• QUICK_REFERENCE.md
• CHANGELOG.md
• etc.

The skill should only contain the information needed for an AI agent to do the job at hand. It should not contain auxiliary context about the process that went into creating it, setup and testing procedures, user-facing documentation, etc. Creating additional documentation files just adds clutter and confusion.

Progressive Disclosure Design Principle

Skills use a three-level loading system to manage context efficiently:

1. Metadata (name + description) - Always in context (~100 words)
2. SKILL.md body - When skill triggers (<5k words)
3. Bundled resources - As needed by Gemini CLI (Unlimited because scripts can be executed without reading into context window)

Progressive Disclosure Patterns

Keep SKILL.md body to the essentials and under 500 lines to minimize context bloat. Split content into separate files when approaching this limit. When splitting out content into other files, it is very important to reference them from SKILL.md and describe clearly when to read them, to ensure the reader of the skill knows they exist and when to use them.

Key principle: When a skill supports multiple variations, frameworks, or options, keep only the core workflow and selection guidance in SKILL.md. Move variant-specific details (patterns, examples, configuration) into separate reference files.

Pattern 1: High-level guide with references

# PDF Processing

## Quick start

Extract text with pdfplumber: [code example]

## Advanced features

- **Form filling**: See [FORMS.md](FORMS.md) for complete guide
- **API reference**: See [REFERENCE.md](REFERENCE.md) for all methods
- **Examples**: See [EXAMPLES.md](EXAMPLES.md

Gemini CLI loads FORMS.md, REFERENCE.md, or EXAMPLES.md only when needed.

Pattern 2: Domain-specific organization

For Skills with multiple domains, organize content by domain to avoid loading irrelevant context:

bigquery-skill/
├── SKILL.md (overview and navigation)
└── reference/
    ├── finance.md (revenue, billing metrics)
    ├── sales.md (opportunities, pipeline)
    ├── product.md (API usage, features)
    └── marketing.md (campaigns, attribution)

When a user asks about sales metrics, Gemini CLI only reads sales.md.

Similarly, for skills supporting multiple frameworks or variants, organize by variant:

cloud-deploy/
├── SKILL.md (workflow + provider selection)
└── references/
    ├── aws.md (AWS deployment patterns)
    ├── gcp.md (GCP deployment patterns)
    └── azure.md (Azure deployment patterns)

When the user chooses AWS, Gemini CLI only reads aws.md.

Pattern 3: Conditional details

Show basic content, link to advanced content:

# CSV Processing

## Basic Analysis

Use pandas for loading and basic queries. See [PANDAS.md](PANDAS.md).

## Advanced Operations

For massive files that exceed memory, see [STREAMING.md](STREAMING.md). For timestamp normalization, see [TIMESTAMPS.md](TIMESTAMPS.md).

Gemini CLI reads REDLINING.md or OOXML.md only when the user needs those features.

Important guidelines:

• Avoid deeply nested references - Keep references one level deep from SKILL.md. All reference files should link directly from SKILL.md.
• Structure longer reference files - For files longer than 100 lines, include a table of contents at the top so Gemini CLI can see the full scope when previewing.

Skill Creation Process

Skill creation involves these steps:

1. Understand the skill with concrete examples
2. Plan reusable skill contents (scripts, references, assets)
3. Initialize the skill (run node init_skill.cjs)
4. Edit the skill (implement resources and write SKILL.md)
5. Package the skill (run node package_skill.cjs)
6. Install and reload the skill
7. Iterate based on real usage

Follow these steps in order, skipping only if there is a clear reason why they are not applicable.

Skill Naming

• Use lowercase letters, digits, and hyphens only; normalize user-provided titles to hyphen-case (e.g., “Plan Mode” -> plan-mode).
• When generating names, generate a name under 64 characters (letters, digits, hyphens).
• Prefer short, verb-led phrases that describe the action.
• Namespace by tool when it improves clarity or triggering (e.g., gh-address-comments, linear-address-issue).
• Name the skill folder exactly after the skill name.

Step 1: Understanding the Skill with Concrete Examples

Skip this step only when the skill’s usage patterns are already clearly understood. It remains valuable even when working with an existing skill.

To create an effective skill, clearly understand concrete examples of how the skill will be used. This understanding can come from either direct user examples or generated examples that are validated with user feedback.

For example, when building an image-editor skill, relevant questions include:

• “What functionality should the image-editor skill support? Editing, rotating, anything else?”
• “Can you give some examples of how this skill would be used?”
• “I can imagine users asking for things like ‘Remove the red-eye from this image’ or ‘Rotate this image’. Are there other ways you imagine this skill being used?”
• “What would a user say that should trigger this skill?”

Avoid interrogation loops: Do not ask more than one or two clarifying questions at a time. Bias toward action: propose a concrete list of features or examples based on your initial understanding, and ask the user to refine them.

Conclude this step when there is a clear sense of the functionality the skill should support.

Step 2: Planning the Reusable Skill Contents

To turn concrete examples into an effective skill, analyze each example by:

1. Considering how to execute on the example from scratch
2. Identifying what scripts, references, and assets would be helpful when executing these workflows repeatedly

Example: When building a pdf-editor skill to handle queries like “Help me rotate this PDF,” the analysis shows:

1. Rotating a PDF requires re-writing the same code each time
2. A scripts/rotate_pdf.cjs script would be helpful to store in the skill

Example: When designing a frontend-webapp-builder skill for queries like “Build me a todo app” or “Build me a dashboard to track my steps,” the analysis shows:

1. Writing a frontend webapp requires the same boilerplate HTML/React each time
2. An assets/hello-world/ template containing the boilerplate HTML/React project files would be helpful to store in the skill

Example: When building a big-query skill to handle queries like “How many users have logged in today?” the analysis shows:

1. Querying BigQuery requires re-discovering the table schemas and relationships each time
2. A references/schema.md file documenting the table schemas would be helpful to store in the skill

To establish the skill’s contents, analyze each concrete example to create a list of the reusable resources to include: scripts, references, and assets.

Step 3: Initializing the Skill

At this point, it is time to actually create the skill.

Skip this step only if the skill being developed already exists, and iteration or packaging is needed. In this case, continue to the next step.

When creating a new skill from scratch, always run the init_skill.cjs script. The script conveniently generates a new template skill directory that automatically includes everything a skill requires, making the skill creation process much more efficient and reliable.

Note: Use the absolute path to the script as provided in the available_resources section.

Usage:

node <path-to-skill-creator>/scripts/init_skill.cjs <skill-name> --path <output-directory>

The script:

• Creates the skill directory at the specified path
• Generates a SKILL.md template with proper frontmatter and TODO placeholders
• Creates example resource directories: scripts/, references/, and assets/
• Adds example files (scripts/example_script.cjs, references/example_reference.md, assets/example_asset.txt) that can be customized or deleted

After initialization, customize or remove the generated SKILL.md and example files as needed.

Step 4: Edit the Skill

When editing the (newly-generated or existing) skill, remember that the skill is being created for another instance of Gemini CLI to use. Include information that would be beneficial and non-obvious to Gemini CLI. Consider what procedural knowledge, domain-specific details, or reusable assets would help another Gemini CLI instance execute these tasks more effectively.

Learn Proven Design Patterns

Consult these helpful guides based on your skill’s needs:

• Multi-step processes: See references/workflows.md for sequential workflows and conditional logic
• Specific output formats or quality standards: See references/output-patterns.md for template and example patterns

These files contain established best practices for effective skill design.

Start with Reusable Skill Contents

To begin implementation, start with the reusable resources identified above: scripts/, references/, and assets/ files. Note that this step may require user input. For example, when implementing a brand-guidelines skill, the user may need to provide brand assets or templates to store in assets/, or documentation to store in references/.

Added scripts must be tested by actually running them to ensure there are no bugs and that the output matches what is expected. If there are many similar scripts, only a representative sample needs to be tested to ensure confidence that they all work while balancing time to completion.

Any example files and directories not needed for the skill should be deleted. The initialization script creates example files in scripts/, references/, and assets/ to demonstrate structure, but most skills won’t need all of them.

Update SKILL.md

Writing Guidelines: Always use imperative/infinitive form.

Frontmatter

Write the YAML frontmatter with name and description:

• name: The skill name
• description: This is the primary triggering mechanism for your skill, and helps Gemini CLI understand when to use the skill.
  • Include both what the Skill does and specific triggers/contexts for when to use it.
  • Must be a single-line string (e.g., description: Data ingestion...). Quotes are optional.
  • Include all “when to use” information here - Not in the body. The body is only loaded after triggering, so “When to Use This Skill” sections in the body are not helpful to Gemini CLI.
  • Example: description: Data ingestion, cleaning, and transformation for tabular data. Use when Gemini CLI needs to work with CSV/TSV files to analyze large datasets, normalize schemas, or merge sources.

Do not include any other fields in YAML frontmatter.

Body

Write instructions for using the skill and its bundled resources.

Step 5: Packaging a Skill

Once development of the skill is complete, it must be packaged into a distributable .skill file that gets shared with the user. The packaging process automatically validates the skill first (checking YAML and ensuring no TODOs remain) to ensure it meets all requirements:

Note: Use the absolute path to the script as provided in the available_resources section.

node <path-to-skill-creator>/scripts/package_skill.cjs <path/to/skill-folder>

Optional output directory specification:

node <path-to-skill-creator>/scripts/package_skill.cjs <path/to/skill-folder> ./dist

The packaging script will:

1. Validate the skill automatically, checking:

   • YAML frontmatter format and required fields
   • Skill naming conventions and directory structure
   • Description completeness and quality
   • File organization and resource references

2. Package the skill if validation passes, creating a .skill file named after the skill (e.g., my-skill.skill) that includes all files and maintains the proper directory structure for distribution. The .skill file is a zip file with a .skill extension.

If validation fails, the script will report the errors and exit without creating a package. Fix any validation errors and run the packaging command again.

Step 6: Installing and Reloading a Skill

Once the skill is packaged into a .skill file, offer to install it for the user. Ask whether they would like to install it locally in the current folder (workspace scope) or at the user level (user scope).

If the user agrees to an installation, perform it immediately using the run_shell_command tool:

• Locally (workspace scope):

  bashCopy

  gemini skills install <path/to/skill-name.skill> --scope workspace

• User level (user scope):

  bashCopy

  gemini skills install <path/to/skill-name.skill> --scope user

Important: After the installation is complete, notify the user that they MUST manually execute the /skills reload command in their interactive Gemini CLI session to enable the new skill. They can then verify the installation by running /skills list.

Note: You (the agent) cannot execute the /skills reload command yourself; it must be done by the user in an interactive instance of Gemini CLI. Do not attempt to run it on their behalf.

Step 7: Iterate

After testing the skill, users may request improvements. Often this happens right after using the skill, with fresh context of how the skill performed.

Iteration workflow:

1. Use the skill on real tasks
2. Notice struggles or inefficiencies
3. Identify how SKILL.md or bundled resources should be updated
4. Implement changes and test again