| name | literature-review |
| description | Systematic literature review methodology and best practices. |
| metadata | {"nanobot": {"always": false}} |
Literature Review Skill
This skill guides you through conducting a systematic literature review on a given topic.
When to Use This Skill
Activate when the user asks for:
- "帮我做一个 XXX 的综述"
- "梳理一下 XXX 方向的发展"
- "总结一下 XXX 领域的关键工作"
- "对比这几篇论文"
- "这个方向的研究脉络是什么"
The SLR Workflow
Phase 1: Scope Definition (与用户确认)
Research Question Formulation
Use PICO framework for clinical/empirical topics:
- Population: What domain/system?
- Intervention: What method/approach?
- Comparison: Compared to what baseline?
- Outcome: What metrics/results?
Or use simpler framework for methods/topics:
- What is the core problem?
- What are the main approaches?
- What are the evaluation benchmarks?
- What are the open challenges?
Inclusion/Exclusion Criteria
Define BEFORE searching:
- Date range (e.g., 2020-2025)
- Publication venues (e.g., top conferences only)
- Paper types (empirical vs. theoretical vs. survey)
- Language (English)
Phase 2: Systematic Search
Multi-Source Search Strategy
- Keyword-based search on arXiv (use
academic_search)
- Seed paper snowballing: Start with 2-3 known key papers, use
citation_graph to find ancestors and descendants
- Author tracking: Identify prolific authors in the area, search their recent work
- Venue filtering: Check recent proceedings of relevant conferences
Search Iteration
- Round 1: Broad search → collect candidate papers
- Round 2: Read abstracts → apply inclusion criteria → keep relevant papers
- Round 3: Full-text reading → extract key information
Phase 3: Information Extraction
For each included paper, extract:
### Paper N
- **Citation**: [Author et al., Year]
- **Problem**: [What problem does it solve?]
- **Approach**: [Method in 1-2 sentences]
- **Key Result**: [Main finding with numbers]
- **Relation to others**: [How it compares to prior work]
Phase 4: Synthesis & Taxonomy
Build a Taxonomy
Organize papers by:
- Method type: e.g., "基于微调的方法", "基于提示的方法", "基于检索的方法"
- Problem dimension: e.g., "效率优化", "效果提升", "可解释性"
- Timeline: Chronological evolution
Identify Patterns
- Convergence: Are different approaches converging on similar solutions?
- Divergence: Are there competing paradigms?
- Gaps: What important problems are understudied?
Generate Insights
Use insight_generator tool if available, or manually synthesize:
- What is the state-of-the-art?
- What are the trade-offs between approaches?
- What are the most promising future directions?
Phase 5: Output Format
# Literature Review: [Topic]
## 1. 研究背景与问题定义
[Why this topic matters, formal problem statement]
## 2. 方法分类与演进
### 2.1 [Category A]
- [Paper 1]: [One-line summary]
- [Paper 2]: [One-line summary]
### 2.2 [Category B]
...
## 3. 关键对比
| Method | Key Idea | Strength | Weakness | Best Result |
|--------|----------|----------|----------|-------------|
| ... | ... | ... | ... | ... |
## 4. 开放问题与未来方向
1. [Open problem 1]
2. [Open problem 2]
## 5. 推荐阅读清单
[Top 5-10 papers, ordered by importance for newcomers]
Quality Control
- Coverage check: Did you find the seminal/foundational papers?
- Balance check: Are you over-representing one approach?
- Recency check: Are recent important papers included (last 1-2 years)?
- Bias check: Did you only search one source? Did you miss non-English work?
Common Mistakes
- ❌ Starting with a narrow search → missing foundational work
- ❌ Including too many papers → shallow analysis
- ❌ Being descriptive only → missing critical analysis
- ❌ Ignoring negative results → over-optimistic view of the field