search

name	search
description	Search academic papers in the local autor knowledge base. Uses auditable node-level FTS5 evidence search, evidence bundles, author search, and top-cited ranking. Use when the user wants to find papers, look up literature, search by author, or explore research topics.

Literature Search

Search for papers in the local library. Defaults to auditable node-level FTS5 retrieval over metadata plus paper.md chunks. For complex questions, generate a bundle with trace/verify artifacts.

Execution Logic

1. Parse user input and determine the search mode:

   • If the user explicitly asks for "semantic search", "vector search", or "vsearch" → explain that this project no longer has vector search and use search or research
   • If the user explicitly asks for "keyword search", "full-text search", "FTS", or ordinary paper search → use search
   • If the user asks a question that needs answer-time provenance → use research
   • If the user explicitly wants to search by author (e.g. "find papers by X", "papers published by X") → use search-author
   • If the user asks to sort by citation count (e.g. "most cited", "classic papers", "top cited") → use top-cited
   • Default: use search for paper retrieval, or research when answer-time provenance is needed.
   • Do not look for an embedding refresh command; vector and FAISS storage has been removed.

2. Extract from user input:

   • Query: what the user wants to find
   • Result count: --top N if specified; otherwise use the default
   • Year filter: --year 2023 (single year), --year 2020-2024 (range), --year 2020- (from year onwards)
   • Journal filter: --journal "Fluid Mechanics" (fuzzy match)
   • Type filter: --type review (fuzzy match; common values: review, journal-article, book-chapter)

3. Run the search command:

Auditable search (default):

autor search "$ARGUMENTS" --top <N> [--year <Y>] [--journal <J>] [--type <T>]

Keyword search:

autor search "<query>" --top <N> [--year <Y>] [--journal <J>] [--type <T>]

Evidence bundle:

autor research "<query>" --top <N> --run-dir workspace/runs/<case>

Author search:

autor search-author "<author name>" --top <N> [--year <Y>] [--journal <J>] [--type <T>]

Top-cited ranking:

autor top-cited --top <N> [--year <Y>] [--journal <J>] [--type <T>]

4. Present the search results to the user. Each result may include evidence snippets with node_id, section, snippet, and ref_path. For research, answer only from the generated bundle and preserve the ## References section.

5. Complex queries: when CLI parameter combinations are insufficient (e.g. filtering by first-author initial, multi-condition intersections, custom sorting), read data/papers/*/meta.json directly in Python. Key JSON fields:

title, authors, first_author, first_author_lastname, year, doi, journal,
abstract, paper_type, citation_count (dict: crossref/semantic_scholar/openalex),
ids, toc, l3

Examples

User says: "Search for papers on turbulent boundary layers." → Run search "turbulent boundary layer"

User says: "Use semantic search to find literature on drag reduction; give me the top 5." → Say vector search is deprecated, then run search "drag reduction" --top 5

User says: "Find papers by Liao Z-M." → Run search-author "Liao"

User says: "What are the most-cited papers in my library?" → Run top-cited --top 10

User says: "Papers on drag reduction after 2020." → Run search "drag reduction" --year 2020-

User says: "Turbulence papers published in JFM." → Run search "turbulence" --journal "Fluid Mechanics"

User says: "Most-cited review articles in my library." → Run top-cited --top 10 --type review