RE-idea-generation

English | 简体中文

An authoritative skills repository for research idea generation, problem discovery, and direction exploration.

This repository focuses on the front end of the research workflow: finding worthwhile problems, reframing vague interests into researchable questions, and generating defensible directions before execution begins. It is not a general end-to-end research toolkit. It is a curated home for skills directly related to ideation, open-question discovery, and early-stage direction selection.

Positioning

• Keep only skills directly related to idea generation, problem discovery, and direction exploration
• Normalize upstream skill snapshots into a portable repository layout
• Serve as the authoritative home for future ideation-related skills in Research-Equality

Included Skills

The skill collection lives under skills/.

Primary Flow Skills

• creative-thinking-for-research: use when the bottleneck is novelty generation, reframing, or escaping local-optimum thinking
• scientific-brainstorming: use when you want an open-ended, domain-aware thought partner before you have a crisp hypothesis
• brainstorming-research-ideas: use when you need diverge → converge → refine with explicit filtering and ranking
• idea-generation: use when you have a topic, partial concept, or runnable scaffold and need scored candidate proposals
• novelty-assessment: use as the last ideation gate before committing to a direction

Branch Skills

• hypothesis-generation: turn observations, anomalies, or preliminary data into competing testable hypotheses with predictions and experiment ideas
• scientific-critical-thinking: interrogate claims, methods, evidence quality, and confounders to expose flaws, hidden assumptions, and genuine open problems

Operating Skills

• ideation: run a structured multi-agent ideation session around a concept seed when one-shot workflows are not enough
• direction-evolution: preserve cross-cycle direction memory and evidence-driven next-step updates after shortlisting, failed validation, or completed stages

See skills/README.md for the skill catalog and routing guidance.

Canonical Flow

To keep overlap low, use this as the default path:

1. Open the space with exactly one starter: creative-thinking-for-research for novelty frameworks, or scientific-brainstorming for conversational scientific exploration.
2. Structure and shortlist with brainstorming-research-ideas.
3. Concretize with idea-generation: use topic-first mode for conceptual inputs, or template-grounded mode when a runnable scaffold already exists.
4. Run novelty-assessment.
5. Preserve what was learned with direction-evolution.

Routing Rules

• Do not use both creative-thinking-for-research and scientific-brainstorming by default. Pick one opener based on whether you need cognitive frameworks or a live scientific thought partner.
• Use idea-generation in one of its two modes rather than splitting proposal generation across multiple skills.
• Use ideation as an orchestration wrapper when a longer, role-separated session is needed. It is not part of the minimal default path.
• Use scientific-critical-thinking as a critique branch before or after proposal generation when the bottleneck is weak reasoning rather than idea quantity.
• Use hypothesis-generation when you start from anomalies, observations, or preliminary results rather than from open-ended ideation.
• Use direction-evolution after ranking, failed validation, or completed stages. It is the post-cycle operating layer, not an initial ideation skill.

Recommended shared artifacts:

• outputs/<topic-slug>/idea_backlog.md for raw candidate ideas
• outputs/<topic-slug>/problem_map.md for tensions, gaps, and stakeholder pain points
• outputs/<topic-slug>/research_log.md for session notes and pivots
• outputs/<topic-slug>/template_gap_map.md for template assumptions, untested seams, and missing evaluation slices
• outputs/<topic-slug>/direction_shortlist.md for ranked final candidates
• outputs/<topic-slug>/novelty_report.json for script-assisted novelty checks and similar-paper logs
• outputs/<topic-slug>/stage_reflection.json for evidence-driven plan updates after a completed stage
• memory/ideation-memory.md for cross-cycle records of promising, failed, and unresolved directions
• ideations/ideation-<slug>-<timestamp>/session/ for multi-agent ideation sessions, briefs, graphs, and continuation lineage

Repository Layout

skills/
  README.md
  brainstorming-research-ideas/
    SKILL.md
  creative-thinking-for-research/
    SKILL.md
  idea-generation/
    SKILL.md
    references/
    scripts/
    templates/
  ideation/
    SKILL.md
    templates/
  direction-evolution/
    SKILL.md
    references/
    templates/
  scientific-brainstorming/
    SKILL.md
    references/
  hypothesis-generation/
    SKILL.md
    references/
  scientific-critical-thinking/
    SKILL.md
    references/
  novelty-assessment/
    SKILL.md
    references/
README.md
README.zh.md
LICENSE

Usage

Prompt examples assume you load or reference the skill directly from this repository:

Read skills/creative-thinking-for-research/SKILL.md and use 2-3 frameworks to generate five novel research directions for long-context software engineering agents.

Read skills/brainstorming-research-ideas/SKILL.md and run the diverge-converge-refine workflow for evaluation gaps in trustworthy code generation.

Read skills/idea-generation/SKILL.md and generate 5 feasible ideas for repository-level problem discovery in trustworthy coding agents, then save a novelty report for the top 2.

Read skills/idea-generation/SKILL.md in template-grounded mode and use the current experiment template to find three under-tested failure modes plus two feasible new research directions grounded in the existing code.

Read skills/direction-evolution/SKILL.md in memory-update mode and update memory/ideation-memory.md from this shortlist, novelty report, and failed validation notes so the next ideation cycle avoids dead ends.

Read skills/direction-evolution/SKILL.md in stage-reflection mode and convert the latest baseline results into a JSON plan update that identifies unmet success signals and the next diagnostic stage.

Read skills/ideation/SKILL.md and run a standard-depth ideation session for "tools that help researchers surface evaluation blind spots", producing a vision doc, idea briefs, and an ideation graph.

Read skills/hypothesis-generation/SKILL.md and turn these anomalous evaluation results into 3 competing, testable hypotheses with differentiating predictions.

Curation Rules

• A skill must directly support idea generation, problem discovery, or research-direction evaluation
• Skills for literature collection, experiments, implementation, paper formatting, or presentation should live in other repositories
• Prefer reusable, portable skills with clear routing boundaries
• Avoid mirroring large upstream repositories wholesale when a focused curation is enough

Provenance

The current skill set was curated from a local source snapshot:

• AI-research-SKILLs/21-research-ideation/brainstorming-research-ideas
• AI-research-SKILLs/21-research-ideation/creative-thinking-for-research
• agent-research-skills/skills/idea-generation
• agent-research-skills/skills/novelty-assessment
• ideation_team_skill/SKILL.md
• claude-scientific-skills/scientific-skills/scientific-brainstorming
• claude-scientific-skills/scientific-skills/hypothesis-generation
• claude-scientific-skills/scientific-skills/scientific-critical-thinking
• ai-scientist/ai_scientist/generate_ideas.py
• ai-scientist/README.md (template structure)
• ai-scientist/templates/nanoGPT_lite/prompt.json
• ai-scientist/templates/nanoGPT_lite/seed_ideas.json
• EvoScientist/EvoScientist/prompts.py
• EvoScientist/EvoScientist/subagent.yaml

Only ideation- and problem-discovery-related skills are retained here. The normalized authoritative versions are the ones under skills/.

About

Open-source skills for research idea generation, problem discovery, and direction exploration in Research-Equality.