| name | lumina-rag-knowledge |
| description | Manage Lumina's retrieval-augmented generation and knowledge ingestion workflow. Use when editing server/src/services/knowledgeBase.ts, server/src/services/pineconeClient.ts, server/src/scripts/knowledgeCli.ts, server/src/models/KnowledgeSource.ts, files under server/knowledge, manifest-based sync inputs, or debugging grounded answers, retrieval relevance, source lifecycle, and inline citations. |
Lumina Rag Knowledge
Overview
Use this skill for retrieval, ingestion, and citation work. The repository treats knowledge management as a CLI-driven flow rather than a UI-managed feature.
Load The Right Reference
- Read
references/cli-workflow.md when updating or using ingestion commands, manifests, or source files.
- Read
references/retrieval-contract.md when changing runtime retrieval, citation formatting, or grounded-answer behavior.
Preserve The Grounding Contract
- Keep answers grounded in indexed sources.
- Preserve inline citations and the associated sources list behavior when changing retrieval or response composition.
- Avoid adding UI-based mutation flows for knowledge content unless the task explicitly requires a product change.
- Prefer updating source content through the CLI and manifest workflows already documented in the repo.
Change Source Lifecycle Carefully
- Preserve stable
externalId usage when designing repeatable updates.
- Keep ingestion, deletion, listing, and sync semantics coherent across CLI commands and model storage.
- If a change affects both retrieval logic and general API behavior, coordinate with
lumina-backend-api.
Finish With Practical Validation
- Run
npm run build in server/ after code changes.
- Use the existing CLI commands for targeted runtime checks when environment variables and external services are available.
- If Pinecone or Gemini credentials are unavailable, state that runtime retrieval was not exercised.