# Devverse RAG Chat

> DevVerse chat, citation, and retrieval workflow. Use when modifying the chat page, the chat API route, citation parsing, local retrieval fallback, Pinecone or Gemini integration, or the article vectorization pipeline.

- Skill: `hoangsonww/devverse-rag-chat` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add hoangsonww/devverse-rag-chat`
- Raw SKILL.md: https://api.skillmd.com/api/skills/hoangsonww/devverse-rag-chat/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: DevOps & Infra
- Author: hoangsonww (https://skillmd.com/u/hoangsonww)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/hoangsonww/devverse-rag-chat

---


# DevVerse RAG And Chat Workflow

1. Preserve the response contract between backend and UI.
The UI expects an `answer` string, a `sources` array, bracket citations in the answer body, and a trailing `Sources:` section that it strips before rendering.

2. Keep citation parsing aligned with UI behavior.
If you change citation formatting, update both `lib/chat-citations.ts` and the chat UI in the same task.

3. Preserve hybrid retrieval unless you are intentionally redesigning it.
`lib/rag.ts` merges Pinecone results with the local lexical fallback from `lib/rag-local.ts`.

4. Read the detailed contract before changing retrieval, prompt format, or vector metadata.
@references/rag-contract.md

5. Assume Pinecone vectors are stale after article or chunking changes until vectorization is rerun.

6. Validate chat changes with both tests and explicit runtime caveats.
Use the `devverse-validate` skill for automated checks, then note any live env checks you could not perform.