| name | survey |
| description | Conduct a systematic literature survey on a research topic. Collects papers, reads key works, synthesizes findings, detects contradictions, identifies gaps, and produces a structured report. |
| always | false |
Literature Survey
How to Conduct a Survey
When the user asks for a literature survey or review:
Step 1: Define Scope
- Clarify the research topic, time range, and focus areas
- Ask user for specific keywords, conferences, or subfields if unclear
Step 2: Search Papers (paper-search skill)
- Search with multiple query variations (e.g., "efficient transformers", "linear attention", "sparse attention")
- Search at least 3 different query angles
- Collect 30-100 papers depending on scope
Step 3: Read Key Papers (paper-read skill)
- Read the top 10-15 most cited papers in detail
- Use
paper-fetch skill to get full text of the most important papers
- Extract: title, authors, year, key contribution, method, results, limitations
Step 4: Classify & Organize
- Group papers by methodology/approach
- Create a taxonomy (e.g., "Sparse Attention", "Linear Attention", "Kernel-based")
- Note which papers cite which (citation relationships)
Step 5: Synthesize Across Papers (cross-paper-synthesis skill)
After collecting and reading papers, synthesize findings:
- Read the
cross-paper-synthesis skill for detailed instructions
- Produce: narrative overview + structured comparison table + temporal evolution
- Identify the "through line" — core insight unifying the field
Step 6: Detect Contradictions (contradiction-detection skill)
Before concluding, check for conflicting results:
- Read the
contradiction-detection skill for detailed instructions
- Flag any papers with conflicting results on the same benchmark or task
- Include a "Contested Claims" section in the final survey if contradictions exist
Step 7: Grade Key Evidence (evidence-grading skill) — optional but recommended
For the 3-5 most important claims the survey will build on:
- Read the
evidence-grading skill for detailed instructions
- Grade each claim A/B+/B/C+/C/D
- Use calibrated hedging language in the survey text
Step 8: Identify Research Gaps (gap-analysis skill)
- Read the
gap-analysis skill for detailed instructions
- Mine limitations and future work sections from key papers
- Cross-reference claimed gaps against recent work to confirm they are still open
- Summarize top 3-5 actionable gaps
Step 9: Write the Survey
Write a structured report with:
- Introduction — what is the field, why it matters
- Taxonomy — classification of approaches
- Method Comparison Table — columns: Paper, Method, Dataset, Key Result, Limitations
- Timeline — how the field evolved (from cross-paper-synthesis)
- Key Findings — common themes and insights
- Contested Claims — contradictions found (from contradiction-detection)
- Open Questions / Gaps — what remains unsolved (from gap-analysis)
- References — BibTeX-ready citations
Step 10: Save Results
- Save the survey to a markdown file in the workspace:
survey_{topic}_{date}.md
- Update MEMORY.md with key papers, findings, and top gaps
- Append to HISTORY.md with summary
Quick vs. Deep Survey
| Mode | Steps | When to use |
|---|
| Quick (Steps 1–4, 9–10) | Search + read + write | User wants fast overview, 10-20 papers |
| Standard (Steps 1–6, 9–10) | + synthesis + contradictions | Default for most surveys, 20-50 papers |
| Deep | All steps | User explicitly asks for thorough analysis; use deep-research skill instead |
Output Format
Save as survey_{topic}_{date}.md in the workspace.