Stats
Actions
Tags
From RAG-assistant
Use this skill when the user wants to ask natural-language questions about their ingested documents, search the knowledge base, or interact with their document corpus conversationally. Activates when the user mentions: asking questions about documents, searching the knowledge base, "what do my docs say about...", "find information about...", chatting with documents, or getting cited answers from ingested content. Also activates when the user wants to explore the document registry conversationally ("what documents do I have?", "list files from source X") or trigger ingestion from the chat interface ("ingest this folder via chat"). Do NOT activate for: running the ingestion pipeline directly (use the doc-ingestion-pipeline skill), or starting/configuring the web UI server.
How this skill is triggered — by the user, by Claude, or both
Slash command
/RAG-assistant:rag-chatThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
The chat interface provides conversational Q&A over ingested documents, powered by
The chat interface provides conversational Q&A over ingested documents, powered by Claude with three tools: semantic search, ingestion, and registry lookup.
1. Documents ingested — run the pipeline first (see doc-ingestion-pipeline skill).
2. API keys set:
export ANTHROPIC_API_KEY="sk-ant-..." # for chat (Claude)
export RAG_EMBEDDING_API_KEY="sk-..." # for search (embeddings)
3. Server running (from project root):
python3 scripts/ui.py
Open http://localhost:7842 → Chat tab.
Claude has three tools available on every request:
| Tool | Triggered by |
|---|---|
search_knowledge_base | Questions about document content |
ingest_documents | "ingest ./path", "add X to knowledge base" |
query_registry | "what documents do you have?", "list files from source X" |
Claude selects the right tool automatically — no commands needed.
What does the onboarding policy say about remote work?
Summarise the key points from the Q1 report.
What documents do I have about SharePoint?
Ingest ./docs/new-policy.pdf
List all documents ingested from the hr-source.
When Claude retrieves documents, a chunk panel appears above the answer showing:
[1], [2], ...) with source name, file path, similarity score (3 decimal places), and a 300-character excerptClaude's answer uses inline citations (e.g. [1], [2]) matching the chunk numbers.
An "Inspect prompt" panel beneath each answer lets you expand and copy the exact augmented prompt sent to the LLM — including the system instruction, all numbered context entries, and the original question.
If the knowledge base is empty, an error message directs you to ingest documents first. If no relevant content is found, Claude says so rather than fabricating an answer.
| Symptom | Fix |
|---|---|
| Chat tab shows "API key not set" | Export ANTHROPIC_API_KEY and restart the server |
| "No relevant information found" | Run ingestion first; verify documents are in .rag-registry.db |
| Streaming cuts off mid-response | Reload the page and retry |
| Ingestion via chat blocked | Another pipeline run is already active; wait for it to finish |
Add an optional [llm] section to .rag-plugin.toml to change model or key env var:
[llm]
model = "claude-sonnet-4-6" # default
llm_key_env = "ANTHROPIC_API_KEY" # default
npx claudepluginhub somtimz/plugins --plugin RAG-assistantBuilds RAG pipelines using LangChain: document loading, recursive text splitting, OpenAI embeddings, and vector stores (Chroma, FAISS, Pinecone).
Queries local document knowledge base using semantic vector search and hybrid retrieval for AI-powered answers with citations via Claude Haiku.