latitude-dev/promptl-skill

Overview

This skill is a concise PromptL syntax guide for writing and reviewing prompts in the Latitude platform. It helps prompt engineers define configurations, messages, variables, conditionals, loops, chains, tools, agents, snippets, and structured outputs. Use it to produce clear, maintainable, and reproducible dynamic prompts for LLMs on Latitude. It focuses on practical syntax examples and common patterns you will use daily.

How this skill works

The skill describes key PromptL constructs and shows minimal working examples for each: YAML configuration blocks to set provider/model/parameters; tagged messages to structure conversation roles; templating for variables, conditionals, and loops; and tags for multi-step chains, tools, agents, and content types. It also covers JSON schema for structured outputs and how to reference reusable prompt snippets. Each example is designed to be copy-paste ready and easy to adapt.

When to use it

• When creating or editing prompts that will run on the Latitude platform
• When configuring LLM provider, model, temperature, or maxSteps
• When adding variables, conditionals, or loops to make prompts dynamic
• When composing multi-step chains or agentic workflows that call tools
• When defining tool parameter schemas or structured JSON outputs

Best practices

• Keep configuration in a top YAML block for clarity and reproducibility
• Use <system>/<user>/<assistant> tags to separate intent and context
• Prefer explicit variable defaults ({{ var || "default" }}) to avoid runtime errors
• Use conditionals and loops to handle optional lists or branching logic cleanly
• Define a strict JSON schema for expected outputs to simplify downstream parsing

Example use cases

• Author a multi-step summarization chain that extracts key facts then formats them as JSON
• Build an agent that searches an internal index and calls a tool for document retrieval
• Create a reusable snippet for company policies and inject it into multiple prompts
• Write a prompt that loops over user-provided items to generate per-item responses
• Configure tools with typed parameter schemas to enable safe function calling