control-literature-search

name	control-literature-search
description	Use for academic literature search in control theory and control engineering, especially when the user asks to find papers, survey a research topic, compare prior work, identify research gaps, or search for scholarly literature. Use paper_search as the primary tool; use current-year arXiv and web_search only as fallback supplements when needed.

Control Literature Search

Use this skill to turn a broad control-theory research question into a traceable candidate paper list. Use paper_search as the primary tool for academic searches. Use current-year arxiv_search only as a preprint supplement, and use web_search/web_fetch only when the primary searches do not produce enough candidates. Do not treat search snippets as final evidence.

Workflow

1. Restate the research target in one sentence.
2. Extract 3-8 core keywords from the user topic.
3. Expand terms with control-domain variants. If the topic is in Chinese, translate to English keywords.
4. Use paper_search as the primary tool for academic paper search. Run 2-3 queries with different keyword combinations.
5. Use web_search only if paper_search plus current-year arXiv search cannot produce enough candidates.
6. Use web_fetch on important result pages when they look authoritative or likely to contain paper metadata.
7. Before presenting final candidate papers as verified, call paper_validate on the candidate list.
8. Produce a candidate paper table with confirmed metadata fields and verification status.
9. Put unverified or mismatch results in a separate "needs verification" section, not in the confirmed paper list.

paper_search Usage

The paper_search tool only supports English queries. Translate Chinese topics to English before searching.

Available parameters:

• query (required): Search keywords
• max_results: Number of results (default 5, max 20)
• sort: Sort order — relevance (default), citations, date
• year_from: Filter papers from this year (inclusive)
• year_to: Filter papers up to this year (inclusive)
• publication_type: Filter by source type — journal, conference, repository
• journal: Limit to one configured control journal, or core for the default core journal set. Common values: core, automatica, ieee_tac, systems_control_letters, ieee_cybernetics, ieee_tcns, ieee_tcst, ieee_tase

Examples:

# Basic search
paper_search(query="event-triggered control multi-agent")

# Filter by year and type
paper_search(query="DoS attack consensus", year_from=2023, year_to=2025, publication_type="journal")

# Search only one journal
paper_search(query="event-triggered control", journal="automatica")

# Sort by citations
paper_search(query="Lyapunov stability networked", sort="citations", max_results=10)

Keyword Expansion

Expand the user's wording before searching. Keep the expansion focused on the actual topic.

Common control variants:

• networked control systems, NCS
• event-triggered control, event-triggered mechanism, self-triggered control
• sampled-data control, periodic event-triggered control
• nonlinear systems, uncertain systems, switched systems, hybrid systems
• adaptive control, robust control, sliding mode control, MPC
• Lyapunov stability, ISS, input-to-state stability, finite-time stability
• LMI, linear matrix inequality, observer-based control
• consensus, multi-agent systems, distributed control
• time delay, packet dropout, quantization, cyber-physical systems

Search Query Strategy

Primary: paper_search queries

Use paper_search for the main academic search. Run 2-3 queries with different keyword combinations:

# Broad topic search
"<topic>" control systems

# Specific method search
"<topic>" Lyapunov stability

# Application-focused
"<topic>" multi-agent systems

The tool automatically returns JSON with title, authors, year, citations, venue, DOI, abstract, OpenAlex source evidence, and verification status when available.

paper_validate Usage

Use paper_validate after collecting candidate papers and before finalizing the paper list. Pass the candidate metadata returned by paper_search or extracted from other sources.

Example:

paper_validate(papers=[
  {"title": "Event-triggered control for networked systems", "doi": "10.xxxx/example", "year": 2024, "venue": "Automatica"}
])

Treat validation statuses as follows:

• confirmed: can appear in the confirmed paper list.
• partial: can appear only with the listed issues preserved.
• unverified: keep as a discovery clue or needs-verification item.
• mismatch: do not present as a confirmed paper.

Fallback: web_search queries

Use these web_search patterns only when paper_search plus current-year arXiv results are insufficient:

IEEE-focused queries:

site:ieeexplore.ieee.org "<topic>" control
"<topic>" "IEEE Transactions on Automatic Control"
"<topic>" "IEEE Transactions on Cybernetics"

Elsevier and Automatica queries:

site:sciencedirect.com "<topic>" control
"<topic>" Automatica
"<topic>" "Systems & Control Letters"

Google Scholar queries:

site:scholar.google.com "<topic>"
"<topic>" "Google Scholar" "cited by"

Source Priority

For control theory, use this priority order:

1. paper_search (OpenAlex) — primary tool for structured metadata, citations, DOI, source evidence, and verification status
2. paper_validate — required validation step before presenting final confirmed papers
3. web_search — fallback supplement only when paper_search plus current-year arXiv search does not produce enough candidates
4. web_fetch — for reading specific publisher pages or author homepages
5. arXiv — for preprints and direct PDF access when other sources don't have full text

arXiv Supplement Rule

Use arXiv only as a current-year preprint supplement unless the user explicitly asks for older preprints.

• If the current year is 2026, call arxiv_search(..., year_from=2026, year_to=2026).
• Do not place arXiv results in the confirmed journal paper list.
• Put arXiv results in a separate "current-year preprints" or "preprint supplement" section.
• If the user asks for a journal set, arXiv results are out-of-scope for that confirmed journal set even when the title is highly relevant.

web_search Fallback Rule

Use web_search only after the primary searches are insufficient:

1. Run paper_search with the requested journal, year, and publication filters.
2. If more candidates are needed, run current-year arxiv_search as a preprint supplement.
3. Only if the combined paper_search and arXiv results are still insufficient, use web_search for IEEE, Elsevier, Google Scholar, or publisher-page discovery.

web_search results are discovery clues by default. Do not place them in the confirmed journal paper list unless their metadata is later verified by paper_validate and they satisfy the requested journal-scope constraints.

Google Scholar pages may be inaccessible or incomplete through web_fetch. Use Google Scholar mainly for discovery signals, citation clues, and title matching; confirm metadata through publisher pages, DOI, Crossref/OpenAlex/Semantic Scholar, arXiv, or author PDFs.

Google Scholar author profile pages, especially URLs like scholar.google.com/citations?user=..., are not paper pages. Do not add an author profile as a candidate paper. If a Scholar author profile snippet mentions a relevant paper title, extract that title and run follow-up exact-title searches such as:

"<paper title>" DOI
"<paper title>" IEEE
"<paper title>" ScienceDirect
"<paper title>" "Google Scholar"
"<paper title>" "cited by"

Only treat a Google Scholar result as a paper-level result when the title/snippet clearly identifies a specific paper. Otherwise label it as an author/profile discovery clue and verify the paper through another source.

Candidate Table

Return a table like this after the first search pass:

| # | Title | Authors | Year | Venue | Source | URL | DOI | Verification |
|---|-------|---------|------|-------|--------|-----|-----|--------------|

Rules:

• Keep the full candidate table columns exactly as shown above. Do not simplify the table by dropping Source, URL, DOI, or Verification.
• Use "unknown" instead of inventing missing authors, years, venues, DOIs, or verification status.
• Do not mark a paper as confirmed unless paper_validate returned confirmed.
• Distinguish "confirmed from source page" from "inferred from snippet".
• Only claim Google Scholar coverage if a query explicitly targeted scholar.google.com or a result URL came from scholar.google.com; otherwise label it as general web search.
• Do not use Google Scholar author profile pages as the primary URL for candidate papers unless no better source is available; if used, mark it as a discovery clue in Source and list missing verification fields.
• Prefer original publisher, DOI, arXiv, or author PDF URLs over aggregator pages.
• If two results look like the same paper, merge them and keep all useful URLs.

Quality Bar

Before summarizing the literature, ensure:

• paper_search was used as the primary search tool with at least 2 different keyword combinations.
• paper_validate was used before finalizing confirmed papers.
• web_search was used only if paper_search plus current-year arXiv results were insufficient.
• If web_search was used, its results were kept as discovery clues unless later validated and in-scope.
• Important candidates were opened with web_fetch when possible.
• Missing metadata is explicitly marked.
• Claims about methods, limitations, or research gaps are not made from snippets alone.

If the search results are thin, say which source family was weak and propose the next query group instead of fabricating a complete review.