kb-query

name: kb-query description: > Answer research questions against the knowledge base using structured lookups across topic files, paper cards, and the index. layer: meta domain: [general] source-project: papers-kb depends-on: [paper-extractor, kb-integrator] tags: [knowledge-base, query, research, sota-lookup, comparison]

KB Query — Research Question Answering

Purpose

Answer research questions by querying the knowledge base. This is a read-only skill — it searches across topic files, paper cards, the index, and the registry but never modifies them.

When to Use

Invoke this skill when the user asks:

• "What's the state of the art in X?"
• "Compare paper A vs paper B"
• "What are the open problems in X?"
• "Which papers use method Y?"
• "What's the timeline of progress in X?"

Not for: Adding papers (use paper-extractor + kb-integrator), or discovering new papers (use paper-discoverer).

Query Types

1. SoTA Lookup

Question pattern: "What's the state of the art in ?"

Procedure:

1. Read kb/config/taxonomy.md to map the question to domain tags
2. Read the relevant topic file(s) in kb/topics/
3. Return the "State of the Art" section with the SoTA paper's full contribution summary
4. Include the results table from the SoTA paper's card

2. Paper Comparison

Question pattern: "Compare vs "

Procedure:

1. Read both paper cards from kb/papers/
2. Build a comparison table:
   • Method approach
   • Benchmarks evaluated
   • Key results (side-by-side where benchmarks overlap)
   • Strengths and limitations of each
3. Note which is classified as SoTA/incremental/etc.

3. Gap Analysis

Question pattern: "What are the open problems in ?"

Procedure:

1. Read the relevant topic file(s)
2. Return the "Open Problems" section
3. Cross-reference with paper cards' "Limitations" sections for additional gaps
4. Check if any open problems have been addressed by recent papers (compare dates)

4. Method Search

Question pattern: "Which papers use <method/technique>?"

Procedure:

1. Grep across kb/papers/*.md for the method name in "Method Summary" sections
2. Grep across kb/topics/*.md in "Methods Landscape" tables
3. Return a list of papers with brief descriptions of how they use the method

5. Timeline Query

Question pattern: "What's the timeline of progress in ?"

Procedure:

1. Read the relevant topic file
2. Read kb/timeline.md and filter entries for the topic's domain tags
3. Read paper cards sorted by year
4. Build a chronological narrative:
   • Year by year, what papers were published and what they contributed
   • Mark SoTA transitions explicitly

6. Coverage Query

Question pattern: "What does the KB cover?" / "How many papers do we have on X?"

Procedure:

1. Read kb/stats.md for high-level metrics
2. Read kb/index.md for the full paper list
3. Filter by domain tags if a specific topic is requested
4. Report: total papers, breakdown by domain, breakdown by novelty classification

7. Free-Form Query

For questions that don't match the above patterns:

1. Grep across all KB files (kb/**/*.md) for relevant keywords
2. Read the most relevant matches
3. Synthesize an answer citing specific papers and topic file sections

Response Format

All responses should:

• Cite specific papers by their card ID (e.g., kong2022-marsim)
• Link to paper cards: kb/papers/<id>.md
• Include quantitative results where available
• Note the novelty classification of cited papers
• Flag any gaps in KB coverage relevant to the question

Example Queries and Expected Responses

Example: "What's the best drone simulator?"

1. Read kb/topics/drone-simulation.md
2. Find SoTA section → identify the leading simulator
3. Return: simulator name, key features, benchmark results, how it compares to alternatives
4. Note: "This is based on N papers in the KB covering drone simulation"

Example: "Compare MARSIM and OmniDrones"

1. Read kb/papers/kong2022-marsim.md and kb/papers/xu2024-omnidrones.md
2. Compare: simulation type (point-realistic vs physics-based), supported features, performance metrics
3. Note: different use cases — MARSIM for lightweight testing, OmniDrones for RL training