baml-codegen

What This Skill Does

The baml-codegen skill provides a specialized interface for generating high-fidelity, type-safe BAML (Boundary AI Markup Language) code. It functions as an automated architect for your LLM-integrated applications, handling everything from data model definitions to complex agent workflows. By interacting with this skill, developers can translate natural language requirements into robust .baml files that define how data is extracted, classified, or retrieved. It bridges the gap between unstructured LLM outputs and structured, type-safe code for Python, TypeScript, Ruby, and Go. The skill leverages real-time pattern matching via MCP to ensure your implementation adheres to BoundaryML best practices, resulting in code that is resilient, performant, and ready for production.

Installation

To integrate this skill into your environment, use the OpenClaw command-line interface: clawhub install openclaw/skills/skills/killerapp/baml-codegen Ensure you have the baml-cli installed in your project path, as the generated files rely on the local compiler to produce the necessary client SDKs.

Use Cases

• Structured Data Extraction: Automatically map unstructured text, PDFs, or multimodal inputs (images/audio) into validated JSON schemas.
• Multi-Step Agent Workflows: Build complex reasoning chains for agents using LangGraph integrations with explicit state transition schemas.
• Resilient LLM Pipelines: Implement automatic retry policies, fallback models (e.g., fast/cheap vs. slow/reliable), and input/output validation for production stability.
• RAG Architectures: Define schema-based retrieval patterns that enforce citation requirements and source accuracy.

Example Prompts

1. "Generate a BAML class for a Customer Support Ticket with fields for sentiment, urgency, and category. Include an enum for priority and a function that uses GPT-4o for classification."
2. "I need a multimodal extraction function that takes an image of an invoice and returns a total amount as a float and a list of line items. Use a fallback client strategy."
3. "Help me refactor my BAML file to implement a retry policy for my OpenAI client and add an assertion to ensure the total price is always greater than zero."

Tips & Limitations

• Schema First: Always prioritize your data model definitions in baml_src/. Think of the schema as the prompt itself; BAML's compiler handles the complex string-to-type parsing.
• Immutability of Client Files: Never manually edit the files inside baml_client/. These are automatically generated and will be overwritten during the build process.
• Iterative Testing: Utilize baml-cli test frequently during development to validate your prompts against edge cases without deploying your entire application stack.
• Context Awareness: The skill works best when provided with the specific target language (e.g., Python or TypeScript) and the intended orchestration framework (e.g., FastAPI, Next.js).