Agent-Skills.md

Agent Skills: Query Papers via RAG

Query papers using RAG (PaperQA2 or LEANN). Use when user needs synthesized answers from papers, asks "what does paper X say about Y", or needs cited responses.

UncategorizedID: hummat/paperpipe/papi-ask

Repository

hummat/paperpipe
hummatLicense: MIT
81

Install this agent skill to your local

pnpm dlx add-skill https://github.com/hummat/paperpipe/tree/HEAD/skills/papi-ask

Skill Files

Browse the full folder contents for papi-ask.

Download Skill

Loading file tree…

skills/papi-ask/SKILL.md

Skill Metadata

Name
papi-ask
Description
Query papers using RAG (PaperQA2 or LEANN). Use when user needs synthesized answers from papers, asks "what does paper X say about Y", or needs cited responses.

Query Papers via RAG

Use papi ask for questions requiring synthesis across papers or cited answers.

Cost-Aware Retrieval

Before using RAG, consider cheaper alternatives:

  1. Exact match: papi search --rg "query" — fast, no LLM
  2. Ranked search: papi search "query" — BM25 ranking
  3. Direct read: papi show <paper> -l eq|tex|summary — if you know the paper

Use papi ask only when:

  • User explicitly requests RAG/synthesis
  • Question spans multiple papers
  • Search/show cannot answer

Commands

# PaperQA2 (default) — full RAG with citations
papi ask "question"

# LEANN — faster semantic search + LLM
papi ask "question" --backend leann

# Structured output for programmatic use
papi ask "question" --format evidence-blocks

MCP Tools (if available)

For quick retrieval without full RAG:

  • leann_search(index_name, query, top_k) — fast semantic search
  • retrieve_chunks(query, index_name, k) — PaperQA2 chunks with citations

Check available indexes: leann_list() or list_pqa_indexes()

Output

RAG answers include:

  • Synthesized response
  • Citations with page/section references
  • Confidence indicators

For general CLI commands, see /papi.