AutoRAG-Research: Skill for Claude Code

Create Generation Plugin

Workflow

1. Scaffold

autorag-research plugin create my_rag --type=generation

Read the generated pipeline.py, pyproject.toml, YAML config, and test file to understand the structure.

2. Implement

For the shared pipeline implementation and testing rules, read:

• ai_instructions/pipeline_implementer.md
• ai_instructions/pipeline_test_writer.md
• ai_instructions/pipeline_architecture_mapper.md

Implement the _generate(query_id, top_k) method. This is where your RAG strategy lives.

Available attributes inside the pipeline:

• self._llm — LangChain BaseLanguageModel (use await self._llm.ainvoke(prompt))
• self._retrieval_pipeline — composed retrieval pipeline (use await self._retrieval_pipeline._retrieve_by_id(query_id, top_k))
• self._service — GenerationPipelineService (use self._service.get_chunk_contents(chunk_ids), self._get_query_text(query_id))

Must return a GenerationResult(text=...) (from autorag_research.orm.service.generation_pipeline).

DO NOT add your own asyncio.gather, asyncio.Semaphore, or any concurrency control. The base pipeline's run() already handles parallel execution of all queries via run_with_concurrency_limit() (semaphore + gather), controlled by the max_concurrency config parameter. Your _generate method is called once per single query — just implement the retrieve-and-generate logic for that one query.

Custom parameters: Add fields to your config class and pass them via get_pipeline_kwargs() → accept them in the pipeline constructor.

Inherited config fields (from BaseGenerationPipelineConfig):

• llm — LLM model string (auto-converted to LangChain model instance)
• retrieval_pipeline_name — name of the retrieval pipeline to compose with (Executor injects it)

3. Write tests and install

Use langchain_core.language_models.FakeListLLM to mock the LLM in tests.

cd my_rag_plugin
pip install -e .   # or: uv pip install -e .
cd .. && autorag-research plugin sync