creating-skills_skill

Overview

This skill guides authors to design and ship effective Letta Code skills that extend the agent with domain workflows, integrations, and reusable resources. It focuses on brevity, progressive disclosure, and packaging reusable scripts, references, and assets so agents remain token-efficient and predictable. Use it to create new skills or improve existing ones for production use.

How this skill works

The skill explains what a well-structured skill contains, how to choose what goes into the small manifest vs external resources, and how to organize scripts, reference documents, and output assets. It prescribes a lightweight loading strategy so the agent only reads detailed references when they're needed, and recommends patterns for domain separation and conditional loading. It also gives a step-by-step creation process from discovery to packaging and iteration.

When to use it

• Creating a new skill to add domain knowledge, workflows, or API integrations to Letta Code
• Refactoring an existing skill to reduce token usage and improve discoverability
• Deciding whether to include a script, reference document, or asset in a skill
• Designing a skill that must support multiple variants or frameworks
• Preparing a skill for deterministic or repeatable operations that benefit from executable scripts

Best practices

• Keep the manifest concise—only include what the agent must always see
• Use external reference files for large or variant-specific details and load them conditionally
• Choose degrees of freedom (high/medium/low) based on task fragility and need for consistency
• Provide executable scripts for deterministic operations and assets for final outputs
• Avoid adding unrelated documentation files; include only resources that directly support the skill
• Organize references one level deep and include a table of contents for long files

Example use cases

• A PDF processing skill that bundles a rotate script and a reference for advanced OCR options
• A database-querying skill that stores table schemas in reference files and sample queries in scripts
• A frontend boilerplate skill that provides an assets folder with a starter React app for reuse
• A compliance skill that keeps policies in references and provides small scripts to validate documents
• A deployment skill that offers provider-specific guides as separate reference files loaded on demand