RE-research-design

English | 简体中文

An authoritative skills repository for research design, method formalization, experiment planning, and evaluation protocol drafting.

This repository focuses on the core upstream research-design workflow: idea generation, novelty assessment, research planning, atomic method decomposition, algorithm specification, experiment design, and statistical analysis. It is not a general end-to-end research-paper toolkit. It is a curated home for skills directly related to designing what to study, how to study it, and how to evaluate it.

Positioning

• Keep only skills related to research ideation, methodology design, experiment planning, and evaluation design
• Normalize everything into a portable repository layout without depending on one author's local machine
• Serve as the authoritative home for future research-design-related skills in Research-Equality
• Keep literature-discovery skills in RE-literature-discovery, and keep paper-writing / submission skills outside this repository

Included Skills

The skill collection lives under skills/.

Strategy and Planning Skills

• brainstorming-research-ideas: structured ideation frameworks for exploring new directions
• creative-thinking-for-research: deeper cognitive creativity frameworks for novel idea generation
• idea-generation: brainstorm and refine candidate research directions
• scientific-brainstorming: conversational scientific ideation and reframing
• trend-grounded-opportunity-finding: derive ranked opportunities from trend and gap signals
• template-grounded-idea-generation: generate ideas that are feasible inside an existing experiment harness
• hypothesis-generation: turn observations and early evidence into testable hypotheses
• novelty-assessment: stress-test whether an idea is actually novel
• research-planning: turn an idea into a concrete research and delivery plan
• success-signal-experiment-planning: turn a direction into staged work with explicit evidence thresholds
• research-template-design: turn a problem family into a stable experiment template
• research-memory-evolution: distill durable lessons from ideation and experiment cycles

Method Formalization Skills

• atomic-decomposition: break a method into traceable conceptual atoms
• math-reasoning: formalize assumptions, notation, derivations, and tests
• algorithm-design: write pseudocode and architecture diagrams

Evaluation Design Skills

• experiment-design: build staged experiment plans, ablations, and metric choices
• condition-role-experiment-design: express experiments with portable reference / proposed / variant conditions
• baseline-anchored-experiment-design: plan limited run sequences relative to an existing baseline
• data-analysis: run statistical summaries, comparisons, and significance checks
• scientific-critical-thinking: critique evidence quality, bias risk, and design validity
• stage-reflection-plan-updates: revise remaining stages after each meaningful checkpoint
• pivot-refine-research-decision: choose whether to proceed, refine, or pivot after early results

See skills/README.md for the catalog and routing rules.

Skill Routing

To avoid overlap, the repository uses one canonical flow plus a small number of explicit forks:

1. Choose one ideation entrypoint: scientific-brainstorming for fuzzy exploration brainstorming-research-ideas for structured divergence creative-thinking-for-research for novelty forcing trend-grounded-opportunity-finding for trend-driven prioritization template-grounded-idea-generation for harness-bounded ideation
2. idea-generation to package and rank candidate directions
3. novelty-assessment
4. hypothesis-generation only when observations must become testable claims
5. research-planning
6. success-signal-experiment-planning
7. research-template-design only if you are building a reusable harness
8. atomic-decomposition and / or math-reasoning, then algorithm-design, when the method needs formal grounding
9. experiment-design
10. Choose one experiment specialization if needed: condition-role-experiment-design for portable schemas baseline-anchored-experiment-design for fixed-harness bounded runs
11. scientific-critical-thinking
12. data-analysis
13. stage-reflection-plan-updates
14. pivot-refine-research-decision
15. research-memory-evolution

Deduping rules:

• Do not run multiple ideation skills by default; pick one entry skill, then consolidate with idea-generation.
• research-planning is the canonical top-level plan; success-signal-experiment-planning is the execution-stage refinement of that plan.
• research-template-design is optional and should only be used for reusable harnesses.
• experiment-design is the default planner; the two experiment specializations are opt-in side branches, not required follow-ups.
• stage-reflection-plan-updates is local plan maintenance, while pivot-refine-research-decision is the global go / no-go decision.

Shared artifact conventions:

• outputs/<project-slug>/idea_candidates.json for candidate ideas
• outputs/<project-slug>/opportunity_map.json for trend-grounded ranked opportunities
• outputs/<project-slug>/novelty_report.json for novelty decisions and nearest prior work
• outputs/<project-slug>/hypotheses.md or hypotheses.json for competing hypotheses and predictions
• outputs/<project-slug>/research_plan.json for the primary study plan
• outputs/<project-slug>/success_criteria.md for stage-level goals and evidence thresholds
• outputs/<project-slug>/template_spec.md for reusable harness or template design
• outputs/<project-slug>/method_atoms.json for atomic decomposition outputs
• outputs/<project-slug>/algorithm_spec.md for pseudocode and architecture notes
• outputs/<project-slug>/experiment_design.json for staged experiment plans
• outputs/<project-slug>/condition_plan.yaml for domain-agnostic condition-role experiment plans
• outputs/<project-slug>/run_plan.md for baseline-anchored bounded run sequences
• outputs/<project-slug>/critical_review.md for design-risk, bias, and evidence-quality audits
• outputs/<project-slug>/analysis/ for statistical summaries and comparison reports
• outputs/<project-slug>/plan_update.json for reflection-driven stage updates
• outputs/<project-slug>/research_decision.md for proceed / refine / pivot decisions
• outputs/<project-slug>/memory/ for ideation and experiment memory distilled from prior cycles

Repository Layout

skills/
  algorithm-design/
  atomic-decomposition/
  brainstorming-research-ideas/
  creative-thinking-for-research/
  data-analysis/
  experiment-design/
  condition-role-experiment-design/
  baseline-anchored-experiment-design/
  hypothesis-generation/
  idea-generation/
  math-reasoning/
  novelty-assessment/
  pivot-refine-research-decision/
  research-memory-evolution/
  research-planning/
  research-template-design/
  scientific-brainstorming/
  scientific-critical-thinking/
  stage-reflection-plan-updates/
  success-signal-experiment-planning/
  template-grounded-idea-generation/
  trend-grounded-opportunity-finding/

Usage

Command examples assume you run them from the repository root.

python skills/idea-generation/scripts/novelty_check.py \
  --idea "Adaptive attention head pruning via gradient-guided importance" \
  --max-rounds 5

python skills/experiment-design/scripts/design_experiments.py \
  --method "contrastive learning with curriculum negatives" \
  --task classification \
  --format markdown

python skills/data-analysis/scripts/stat_summary.py \
  --input results.csv \
  --compare method \
  --metric accuracy \
  --output outputs/demo/analysis/summary.json

Recommended environment:

• Python 3.10+
• Optional environment variable: SEMANTIC_SCHOLAR_API_KEY
• Optional dependencies: numpy, scipy

Curation Rules

• A skill must directly support research ideation, method formalization, experiment planning, or evaluation design
• Literature discovery, citation management, and survey writing should live in RE-literature-discovery
• Implementation, paper formatting, compilation, rebuttal, and slide-generation skills should not live here
• Prefer skills that are scriptable, reusable, and auditable

Provenance

The current skill set was curated from local source snapshots including agent-research-skills/, AI-research-SKILLs/, claude-scientific-skills/, ai-scientist/, AutoResearchClaw/, and EvoScientist/.

Only research-design-related skills are retained here. Overlapping or downstream capabilities were intentionally excluded so the repository stays focused on upstream study design rather than the full paper lifecycle. The normalized authoritative version is the one under skills/.