@wentorai/research-plugins 1.0.0 → 1.2.0

package/skills/research/paper-review/paper-critique-framework/SKILL.md

@@ -0,0 +1,181 @@
+---
+name: paper-critique-framework
+description: "Structured framework for writing peer review reports and paper critiques"
+metadata:
+  openclaw:
+    emoji: "📋"
+    category: "research"
+    subcategory: "paper-review"
+    keywords: ["peer review", "paper critique", "referee report", "academic review", "manuscript evaluation", "constructive feedback"]
+    source: "https://clawhub.ai/peer-reviewer"
+---
+
+# Paper Critique and Peer Review Framework
+
+## Overview
+
+Writing constructive peer reviews is a core academic skill. This framework provides a systematic approach to evaluating manuscripts — from initial read-through to the final referee report. It covers what reviewers should assess, how to structure feedback, and how to calibrate between different review outcomes (accept, revise, reject). Applicable to conference papers, journal articles, and internal lab reviews.
+
+## The Three-Pass Review Method
+
+### Pass 1: Orientation (15-20 minutes)
+
+Read only these elements:
+- Title, abstract, and keywords
+- Introduction (first and last paragraphs)
+- Section headings and figure captions
+- Conclusion
+
+After Pass 1, answer:
+```
+□ What is the main claim?
+□ What type of contribution? (empirical, theoretical, system, survey)
+□ Is it within the venue's scope?
+□ Does the abstract accurately represent the content?
+□ Initial impression: novel or incremental?
+```
+
+### Pass 2: Detailed Read (60-90 minutes)
+
+Read the full paper. Annotate as you go:
+
+```markdown
+Annotation symbols:
+  ? = I don't understand this
+  ! = This is interesting / strong point
+  X = I disagree / see a problem
+  → = This needs more evidence or justification
+  ≈ = This is similar to [existing work] — check novelty
+```
+
+Focus on:
+- **Claims vs. evidence**: Is every major claim supported by data?
+- **Methodology**: Are the methods appropriate for the research question?
+- **Experimental design**: Are baselines fair? Are ablations sufficient?
+- **Figures and tables**: Do they support the narrative? Are they readable?
+- **Writing quality**: Is it clear, concise, and well-organized?
+
+### Pass 3: Verification (30-60 minutes)
+
+For papers you're seriously evaluating:
+- Check key references — do they say what the authors claim?
+- Verify mathematical derivations (spot-check, not exhaustive)
+- Examine statistical claims (p-values, confidence intervals, effect sizes)
+- Check for cherry-picking in results (only best runs? selected metrics?)
+- Look for missing baselines that should have been compared
+
+## Review Report Structure
+
+```markdown
+## Summary (3-5 sentences)
+[Describe what the paper does, the approach, and the main finding.
+ Demonstrate that you understood the paper.]
+
+## Strengths (bulleted list)
+- S1: [Specific strength with evidence from the paper]
+- S2: [Another strength]
+- S3: [Another strength]
+
+## Weaknesses (bulleted list, ordered by severity)
+- W1 (Major): [Specific weakness + why it matters + suggestion to fix]
+- W2 (Major): [Another major weakness]
+- W3 (Minor): [A less critical issue]
+- W4 (Minor): [Another minor issue]
+
+## Questions for Authors
+- Q1: [Something you'd like clarified]
+- Q2: [A concern that the authors might be able to address]
+
+## Detailed Comments
+[Page/line-specific comments, typos, suggestions]
+
+## Overall Assessment
+Recommendation: [Strong Accept / Accept / Weak Accept / Borderline /
+                  Weak Reject / Reject / Strong Reject]
+Confidence: [High / Medium / Low]
+```
+
+## Assessment Criteria by Dimension
+
+| Dimension | Questions to Ask | Weight |
+|-----------|-----------------|--------|
+| **Novelty** | Is the idea new? Is the contribution beyond incremental? | High |
+| **Significance** | Would this matter to the community? Does it advance the field? | High |
+| **Soundness** | Are the methods correct? Are conclusions supported? | High |
+| **Clarity** | Is it well-written? Can it be understood and reproduced? | Medium |
+| **Completeness** | Are related works covered? Are experiments thorough? | Medium |
+| **Reproducibility** | Could someone replicate this? Code/data available? | Medium |
+
+### Calibration Guide
+
+```
+Strong Accept: Significant contribution, technically sound, well-written.
+  Would be a highlight of the venue.
+
+Accept: Solid contribution with minor issues. Advances the field.
+  Worth publishing as-is or with minor revisions.
+
+Weak Accept: Has merit but notable weaknesses. Contribution is real but modest.
+  Borderline for this venue; would be accepted at a less selective venue.
+
+Borderline: Equal arguments for and against. Significant weaknesses offset
+  by some novelty. Depends on other reviews.
+
+Weak Reject: Interesting direction but fundamental issues not addressed.
+  Major revisions needed that likely require a new submission cycle.
+
+Reject: Significant problems in novelty, soundness, or relevance.
+  Not suitable for this venue even with revisions.
+
+Strong Reject: Fundamental flaws. Clearly below threshold.
+```
+
+## Common Review Pitfalls to Avoid
+
+| Pitfall | Better Approach |
+|---------|----------------|
+| "The writing needs improvement" (vague) | Give 2-3 specific examples with suggested fixes |
+| Rejecting for not solving YOUR problem | Evaluate the paper on its own stated goals |
+| Demanding impossible experiments | Suggest feasible improvements within scope |
+| Ignoring supplementary material | Check appendix — authors may have addressed your concern |
+| Being harsh without being constructive | Every weakness should include a suggestion for improvement |
+| Reviewing too quickly | Block dedicated time; a rushed review harms both authors and science |
+| Citing only your own work as "missing" | Only cite if genuinely relevant, not self-promotion |
+
+## Reviewing Different Paper Types
+
+### Empirical Papers
+- Are datasets described completely? (Size, source, splits, preprocessing)
+- Are baselines appropriate and fairly tuned?
+- Statistical significance: error bars, multiple runs, significance tests
+- Ablation studies: which components contribute to the gain?
+
+### Systems Papers
+- Is the system actually built and evaluated (not just designed)?
+- Performance at scale: latency, throughput, resource usage
+- Comparison with existing systems (not just algorithms)
+- Deployment experience or case studies
+
+### Survey/Position Papers
+- Coverage: are major relevant works included?
+- Taxonomy: is the organizational framework useful and complete?
+- Analysis: does the survey add value beyond listing papers?
+- Recency: are the most recent advances covered?
+
+## Ethics Considerations
+
+```markdown
+During review, check:
+□ Are human subjects involved? Was IRB/ethics approval obtained?
+□ Are there potential harms from the technology described?
+□ Is the data collection ethical? (Consent, privacy, bias)
+□ Are dual-use concerns addressed? (Misuse potential)
+□ Are limitations and societal implications discussed?
+```
+
+## References
+
+- Smith, R. (2006). "Peer review: a flawed process at the heart of science and journals." *Journal of the Royal Society of Medicine*, 99(4), 178-182.
+- [NeurIPS Reviewer Guidelines](https://neurips.cc/Conferences/2025/ReviewerGuidelines)
+- [ACL Reviewer Tutorial](https://aclrollingreview.org/reviewertutorial)
+- Cormode, G. (2009). "How NOT to review a paper." *ACM SIGMOD Record*, 37(4), 100-104.

package/skills/research/paper-review/paper-digest-guide/SKILL.md

@@ -0,0 +1,240 @@
+---
+name: paper-digest-guide
+description: "Fetch paper and spawn sub-agents to read citations recursively"
+metadata:
+  openclaw:
+    emoji: "🔗"
+    category: "research"
+    subcategory: "paper-review"
+    keywords: ["citation analysis", "recursive reading", "paper digestion", "citation graph", "reference mining", "snowball search"]
+    source: "https://github.com/AcademicSkills/paper-digest-guide"
+---
+
+# Paper Digest Guide
+
+A skill for deeply digesting academic papers by recursively following and analyzing their citation networks. Starting from a seed paper, it spawns sub-tasks to read the most important cited references, extracts the intellectual lineage of key claims, and builds a citation-aware knowledge graph that shows how ideas developed and where the research frontier stands.
+
+## Overview
+
+A single paper rarely tells the full story. Key claims rest on cited evidence, methodologies build on prior work, and the significance of findings depends on what came before. The Paper Digest approach addresses this by not just reading a paper but also selectively reading its most important references, and optionally their references, building a multi-level understanding of the intellectual context. This process, often called "citation snowballing" or "reference mining," is one of the most effective ways to deeply understand a research area.
+
+The skill implements a controlled recursive reading strategy with configurable depth (typically 1-2 levels) and breadth (top 5-10 most relevant citations per paper). It produces a citation knowledge graph annotating each reference with its role in the seed paper's argument, along with a synthesis report that traces the evolution of key ideas.
+
+## Citation Digestion Pipeline
+
+### Pipeline Architecture
+
+```
+Seed Paper
+  │
+  ├── Parse references (extract all citations)
+  │
+  ├── Classify citation role for each reference
+  │   ├── Foundational (key theory/framework the paper builds on)
+  │   ├── Methodological (method the paper adopts/adapts)
+  │   ├── Empirical (prior findings the paper extends)
+  │   ├── Contrasting (work the paper disagrees with)
+  │   └── Peripheral (background/context, low priority)
+  │
+  ├── Prioritize: rank by role importance + citation count
+  │
+  ├── Spawn sub-reads for top-N references
+  │   ├── Sub-read 1: [Foundational paper] → structured summary
+  │   ├── Sub-read 2: [Key method paper] → structured summary
+  │   ├── Sub-read 3: [Contrasting paper] → structured summary
+  │   └── ...
+  │
+  ├── (Optional) Depth 2: repeat for sub-read citations
+  │
+  └── Synthesize: build citation knowledge graph + narrative
+```
+
+### Citation Role Classification
+
+```python
+def classify_citation_role(citation_context: str, paper_section: str) -> str:
+    """
+    Classify the role of a citation based on its context in the paper.
+
+    Args:
+        citation_context: The sentence(s) containing the citation
+        paper_section: Which section the citation appears in
+    """
+    # Heuristic classification based on section and language
+    role_indicators = {
+        'foundational': {
+            'sections': ['introduction', 'background', 'related work'],
+            'phrases': ['builds on', 'based on', 'following', 'framework of',
+                       'theory proposed by', 'seminal work']
+        },
+        'methodological': {
+            'sections': ['methods', 'methodology', 'approach'],
+            'phrases': ['we adopt', 'following the approach of', 'using the method',
+                       'as proposed by', 'we extend the method']
+        },
+        'empirical': {
+            'sections': ['introduction', 'results', 'discussion'],
+            'phrases': ['found that', 'showed that', 'demonstrated',
+                       'reported', 'consistent with', 'prior studies']
+        },
+        'contrasting': {
+            'sections': ['introduction', 'related work', 'discussion'],
+            'phrases': ['unlike', 'in contrast to', 'however', 'whereas',
+                       'disagree', 'limitations of', 'fails to']
+        },
+        'peripheral': {
+            'sections': ['introduction'],
+            'phrases': ['for example', 'such as', 'see also', 'e.g.',
+                       'for a review see', 'has been studied']
+        }
+    }
+
+    for role, indicators in role_indicators.items():
+        section_match = paper_section.lower() in [s.lower() for s in indicators['sections']]
+        phrase_match = any(p in citation_context.lower() for p in indicators['phrases'])
+        if section_match and phrase_match:
+            return role
+
+    return 'peripheral'  # default
+```
+
+## Citation Priority Ranking
+
+### Prioritization Criteria
+
+| Criterion | Weight | Rationale |
+|-----------|--------|-----------|
+| Citation role: foundational | 5 | Must understand to grasp the paper |
+| Citation role: contrasting | 4 | Shows alternative perspectives |
+| Citation role: methodological | 4 | Needed to evaluate the approach |
+| Citation role: empirical | 3 | Supports key claims |
+| Cited multiple times in paper | +2 | Indicates higher importance |
+| High citation count (>100) | +1 | Widely recognized work |
+| Recent (<3 years old) | +1 | Current state of the art |
+| Citation role: peripheral | 1 | Low priority for deep reading |
+
+```python
+def rank_citations(citations: list) -> list:
+    """
+    Rank citations by priority for recursive reading.
+    Returns sorted list with top candidates first.
+    """
+    role_weights = {
+        'foundational': 5, 'contrasting': 4, 'methodological': 4,
+        'empirical': 3, 'peripheral': 1
+    }
+
+    for c in citations:
+        score = role_weights.get(c['role'], 1)
+        if c.get('times_cited_in_paper', 1) > 1:
+            score += 2
+        if c.get('global_citations', 0) > 100:
+            score += 1
+        if c.get('year', 0) >= 2023:
+            score += 1
+        c['priority_score'] = score
+
+    return sorted(citations, key=lambda x: x['priority_score'], reverse=True)
+```
+
+## Recursive Reading Configuration
+
+### Depth and Breadth Control
+
+```yaml
+digest_config:
+  # Depth: how many levels of citations to follow
+  max_depth: 2      # 1 = direct citations only, 2 = citations of citations
+
+  # Breadth: how many citations to read at each level
+  top_n_per_level:
+    depth_1: 8      # Read top 8 references from seed paper
+    depth_2: 3      # Read top 3 references from each depth-1 paper
+
+  # Focus: which citation roles to prioritize
+  priority_roles: ["foundational", "methodological", "contrasting"]
+
+  # Read depth for sub-papers
+  sub_read_depth: "pass_2"  # pass_1 (survey), pass_2 (comprehension), pass_3 (critical)
+
+  # Termination conditions
+  max_total_papers: 30       # Hard cap on total papers read
+  stop_on_saturation: true   # Stop when no new concepts are found
+```
+
+### Estimated Time and Output
+
+| Depth | Breadth | Papers Read | Time Estimate | Output Size |
+|-------|---------|-------------|---------------|-------------|
+| 1 | Top 5 | ~6 | 1-2 hours | 3-5 page report |
+| 1 | Top 10 | ~11 | 2-4 hours | 5-8 page report |
+| 2 | Top 8 + Top 3 | ~30 | 4-8 hours | 10-15 page report |
+
+## Citation Knowledge Graph
+
+### Building the Graph
+
+```python
+def build_citation_graph(seed_paper: dict, digested_papers: list) -> dict:
+    """
+    Build a knowledge graph from the digested papers.
+    Nodes are papers, edges are citation relationships with roles.
+    """
+    graph = {
+        'nodes': [],
+        'edges': [],
+        'clusters': {}
+    }
+
+    # Add seed paper as root node
+    graph['nodes'].append({
+        'id': seed_paper['doi'],
+        'label': f"{seed_paper['first_author']} ({seed_paper['year']})",
+        'type': 'seed',
+        'summary': seed_paper['main_finding']
+    })
+
+    # Add digested papers and their relationships
+    for paper in digested_papers:
+        graph['nodes'].append({
+            'id': paper['doi'],
+            'label': f"{paper['first_author']} ({paper['year']})",
+            'type': 'reference',
+            'role': paper['citation_role'],
+            'depth': paper['depth_level'],
+            'summary': paper['main_finding']
+        })
+        graph['edges'].append({
+            'source': paper['cited_by_doi'],
+            'target': paper['doi'],
+            'role': paper['citation_role'],
+            'context': paper['citation_context']
+        })
+
+    return graph
+```
+
+### Interpreting the Graph
+
+The resulting knowledge graph reveals:
+
+- **Intellectual lineage**: The chain of foundational works that led to the seed paper.
+- **Methodological genealogy**: Which methods were inherited and from where.
+- **Debate structure**: Papers on opposing sides of a disagreement, traced through contrasting citations.
+- **Convergence points**: Papers cited by multiple branches, indicating central concepts.
+- **Research frontier**: Recent papers at the leaves of the graph, indicating current directions.
+
+## Best Practices
+
+- Start with depth 1 and increase only if the topic requires deeper historical context.
+- Prioritize contrasting citations; they provide the most critical perspective on the seed paper.
+- When a citation cannot be accessed (paywall, unavailable), note it as a gap rather than skipping silently.
+- Use the citation knowledge graph to identify your own paper's position in the literature.
+- Save the full graph for reuse; it can seed future literature reviews on related topics.
+- Cross-check that the seed paper's characterization of cited work is accurate by reading the originals.
+
+## References
+
+- Wohlin, C. (2014). Guidelines for Snowballing in Systematic Literature Studies. *EASE 2014*.
+- Greenhalgh, T. & Peacock, R. (2005). Effectiveness and Efficiency of Search Methods. *BMJ*, 331, 1064-1065.
+- Chen, C. (2006). CiteSpace II: Detecting and Visualizing Emerging Trends and Transient Patterns. *JASIST*, 57(3), 359-377.

package/skills/research/paper-review/paper-research-assistant/SKILL.md

@@ -0,0 +1,231 @@
+---
+name: paper-research-assistant
+description: "Read papers, generate structured reports, find code and datasets"
+metadata:
+  openclaw:
+    emoji: "📄"
+    category: "research"
+    subcategory: "paper-review"
+    keywords: ["paper reading", "research report", "code extraction", "dataset discovery", "paper analysis", "structured summary"]
+    source: "https://github.com/AcademicSkills/paper-research-assistant"
+---
+
+# Paper Research Assistant
+
+A skill for systematically reading academic papers and generating structured analysis reports. Goes beyond simple summarization by extracting methodology details, identifying associated code repositories and datasets, evaluating reproducibility, and generating actionable summaries suitable for lab meetings, journal clubs, or literature review inclusion.
+
+## Overview
+
+Reading academic papers efficiently is a core skill for researchers, yet many lack a systematic approach. The result is either superficial skimming that misses critical details or time-consuming deep reads of papers that turn out to be marginally relevant. This skill provides a structured reading protocol that adapts its depth to the paper's relevance, extracts standardized metadata, and produces reports that can be shared with collaborators or filed for future reference.
+
+A distinguishing feature is the automated search for associated resources: code repositories, datasets, pre-trained models, and supplementary materials that are often scattered across GitHub, institutional pages, and data repositories. These resources are essential for reproducibility and building upon published work.
+
+## Structured Reading Protocol
+
+### Three-Pass Reading Method
+
+```
+Pass 1: SURVEY (5-10 minutes)
+  Read: Title, abstract, introduction (last paragraph), section headings,
+        figures/tables (captions only), conclusion (first paragraph)
+  Decide: Is this paper relevant enough for a deeper read?
+  Output: One-paragraph relevance assessment
+
+Pass 2: COMPREHENSION (30-60 minutes)
+  Read: Full paper, but skip mathematical derivations and implementation details
+  Focus: What problem? What approach? What results? What limitations?
+  Mark: Key claims, novel contributions, and points of confusion
+  Output: Structured summary (see template below)
+
+Pass 3: CRITICAL ANALYSIS (30-60 minutes, only for highly relevant papers)
+  Read: Every detail including proofs, appendices, supplementary materials
+  Focus: Are the claims justified? Are there hidden assumptions?
+         Could I reproduce this? What would I do differently?
+  Output: Critical evaluation with reproducibility assessment
+```
+
+### Structured Summary Template
+
+```yaml
+paper_summary:
+  title: ""
+  authors: []
+  venue: ""  # journal/conference name
+  year: 0
+  doi: ""
+
+  classification:
+    type: ""  # empirical, theoretical, methodological, review, position
+    domain: ""
+    subdomain: ""
+
+  core_content:
+    problem: |
+      What specific problem does this paper address?
+      Why is it important?
+    approach: |
+      What method/framework/model is proposed?
+      What is novel about the approach?
+    key_results:
+      - result: ""
+        metric: ""
+        value: ""
+        baseline_comparison: ""
+    limitations:
+      - ""
+    future_work:
+      - ""
+
+  methodology:
+    data:
+      datasets_used: []
+      sample_size: ""
+      data_availability: ""  # public, restricted, proprietary
+    method:
+      type: ""  # experimental, observational, simulation, etc.
+      tools: []  # software, libraries, frameworks used
+      reproducibility_score: ""  # high, medium, low
+    evaluation:
+      metrics: []
+      baselines: []
+      statistical_tests: []
+
+  resources:
+    code_repository: ""
+    datasets: []
+    pretrained_models: []
+    supplementary_url: ""
+    demo_url: ""
+
+  assessment:
+    relevance_to_my_work: ""  # high, medium, low
+    quality_rating: ""  # 1-5 scale
+    key_takeaway: ""
+    follow_up_actions: []
+```
+
+## Code and Dataset Discovery
+
+### Finding Associated Resources
+
+```python
+def find_paper_resources(title: str, authors: list, doi: str = None) -> dict:
+    """
+    Search for code, datasets, and other resources associated with a paper.
+
+    Search locations (in priority order):
+    1. Paper itself (check "Code Availability" section)
+    2. Papers With Code (paperswithcode.com)
+    3. GitHub search (title + author name)
+    4. Author institutional pages
+    5. Zenodo / Figshare (for datasets)
+    6. Hugging Face (for models and datasets)
+    """
+    resources = {
+        'code': [],
+        'datasets': [],
+        'models': [],
+        'supplementary': []
+    }
+
+    # Strategy 1: Papers With Code
+    # Search: https://paperswithcode.com/search?q={title}
+    # Returns: code repos, datasets, benchmark results
+
+    # Strategy 2: GitHub search
+    # Query: "{title}" OR "{first_author} {key_method_term}"
+    # Filter: recently updated, has README, has stars
+
+    # Strategy 3: Zenodo/Figshare
+    # Query: DOI or title
+    # Filter: type=dataset
+
+    # Strategy 4: Hugging Face
+    # Query: paper title or model name
+    # Filter: models, datasets
+
+    # Strategy 5: Google search
+    # Query: "{title}" (code OR github OR repository)
+    # Query: "{title}" (dataset OR data OR download)
+
+    return resources
+```
+
+### Reproducibility Assessment
+
+| Factor | Score 3 (High) | Score 2 (Medium) | Score 1 (Low) |
+|--------|---------------|-----------------|--------------|
+| Code availability | Public repo with instructions | Code available on request | No code |
+| Data availability | Public dataset with DOI | Available on request | Proprietary |
+| Method description | Sufficient to reimplement | Missing some details | Vague/incomplete |
+| Hyperparameters | All reported | Key ones reported | Not reported |
+| Environment | Docker/requirements.txt | Software versions listed | Not specified |
+| Random seeds | Fixed and reported | Fixed but not reported | Not controlled |
+
+## Report Generation
+
+### Lab Meeting Presentation Format
+
+```markdown
+## [Paper Title] - [First Author] et al. ([Year])
+
+### Problem
+[2-3 sentences on the problem and why it matters]
+
+### Approach
+[3-4 sentences on the method, emphasizing what is novel]
+
+### Key Results
+- [Result 1 with specific numbers]
+- [Result 2 with specific numbers]
+- [Comparison to best baseline]
+
+### Strengths
+- [Strength 1]
+- [Strength 2]
+
+### Weaknesses / Questions
+- [Weakness 1]
+- [Question for discussion]
+
+### Relevance to Our Work
+[1-2 sentences on how this connects to the lab's research]
+
+### Resources
+- Code: [URL or "not available"]
+- Data: [URL or "not available"]
+```
+
+### Literature Review Entry Format
+
+For inclusion in a systematic literature review, generate:
+
+1. **Bibliographic entry**: Full citation in the target format (APA, Vancouver, etc.).
+2. **Data extraction row**: Structured data for the evidence synthesis table.
+3. **Quality assessment**: Score on the relevant quality assessment tool (CASP, Newcastle-Ottawa, etc.).
+4. **Synthesis note**: How this paper relates to others in the review.
+
+## Batch Processing Workflow
+
+When reading multiple papers on the same topic:
+
+1. **Triage**: Read Pass 1 for all papers. Sort by relevance.
+2. **Prioritize**: Full read (Pass 2+3) only for high-relevance papers.
+3. **Cross-reference**: After reading all papers, build a comparison matrix.
+4. **Synthesize**: Identify points of agreement, disagreement, and gaps.
+5. **File**: Store all structured summaries for future retrieval.
+
+## Best Practices
+
+- Always start with Pass 1. Do not commit to a deep read before assessing relevance.
+- Read the figures and tables early. In empirical papers, they often tell the core story.
+- Note your questions and confusions during reading. These often point to genuine gaps or weaknesses.
+- Search for code and datasets before attempting to reproduce results manually.
+- When a paper cites a finding as established, trace back to the original source and verify.
+- Keep a running log of papers read with one-sentence summaries for quick future reference.
+
+## References
+
+- Keshav, S. (2007). How to Read a Paper. *ACM SIGCOMM Computer Communication Review*, 37(3), 83-84.
+- Pautasso, M. (2013). Ten Simple Rules for Writing a Literature Review. *PLoS Computational Biology*, 9(7).
+- Raff, E. (2019). A Step Toward Quantifying Independently Reproducible Machine Learning Research. *NeurIPS 2019*.