adapters/provider-adapter-map.md

# Provider Adapter Map

## Purpose
This file defines how one Luuna skill is translated to different model providers.

## Universal input
Each adapter receives the same neutral skill definition:

```txt
skill_id
name
purpose
triggers
inputs
outputs
context_needed
steps
validation
install_notes
```

## OpenAI
Best format:
- system or developer instruction
- structured output schema
- tool/function schema when available

Output style:
- strict JSON or Markdown + JSON

## Claude
Best format:
- project instruction
- XML-like sections
- clear role, task, context and output boundaries

Output style:
- Markdown with named sections

## Gemini
Best format:
- short instruction blocks
- JSON-friendly structures
- multimodal context notes when useful

Output style:
- concise Markdown or JSON

## LM Studio
Best format:
- local system prompt
- OpenAI-compatible API assumption
- simple context windows

Output style:
- plain Markdown
- short schemas

## Ollama
Best format:
- Modelfile SYSTEM block
- simple deterministic instructions
- small context footprint

Output style:
- Modelfile plus prompt

## OpenRouter
Best format:
- router-level instruction
- model fallback notes
- provider-specific variants where needed

Output style:
- Markdown + model routing table

## Hugging Face
Best format:
- repo files
- Space README
- Gradio/Streamlit notes
- Transformers pipeline notes

Output style:
- Markdown files and config notes

## Unknown future provider
Best format:
- neutral Markdown
- JSON schema
- minimal assumptions

Output style:
- portable skill spec