luoyu100 Investigate Research
Overview
Conduct a targeted research investigation for computer-science topics. Optimize for usefulness to the user's problem, not for collecting papers only.
Workflow
1. Intake
Ask for exactly two things first:
- The research target or question
- The output directory for the markdown report
If the target is still broad, ask one short follow-up about scope, for example: time range, preferred subarea, or desired depth.
2. Frame the problem
Translate the target into:
- What decision or understanding the user needs
- What would count as a useful answer
- Which subquestions matter most
- Which search terms, aliases, and adjacent concepts should be used
3. Investigate
Prioritize primary and technical sources, especially:
- arXiv and other preprints for current ideas
- IEEE and ACM papers for peer-reviewed work
- Official docs, benchmark papers, and reference implementations
- High-signal surveys, tutorials, and system writeups when they help synthesize the field
Treat non-paper sources as supporting evidence, not replacements for the core literature. Always look for what can help solve the user's problem: methods, constraints, evaluation results, failure modes, implementation details, and open gaps.
4. Synthesize
Write findings in a way that is directly actionable:
- Explain what the topic is and why it matters
- Compare the main approaches or papers
- Call out what works, what fails, and under what conditions
- Identify ideas, techniques, or components that can transfer to the user's problem
- Note limitations, disagreements, and open questions
If sources conflict, prefer primary sources and newer surveys. If evidence is weak, say so explicitly.
5. Write the report
Save the final output as a markdown file in the directory the user provided. Use a clear, searchable filename such as investigation-YYYY-MM-DD-topic.md unless the user specifies another name.
Use this structure:
- Title
- Research target
- Scope and search strategy
- Key findings
- Comparison table if multiple approaches exist
- What is useful for the user's problem
- Risks, gaps, and open questions
- References
Keep citations readable with links or footnotes. Do not dump raw links without context.
Style Rules
- Be concise, but not shallow
- Prefer synthesis over listing
- Tie every major finding back to the target problem
- Avoid overfitting to one source or one paper
- Mention uncertainty when the evidence is incomplete