Research Ideation — Systematic Idea Management & Evaluation

A domain-agnostic system for generating, evaluating, and managing research ideas. Designed for academic researchers in any field who need to systematically explore research directions rather than relying on ad-hoc inspiration.

Core Philosophy

Ideas are not random sparks — they can be systematically generated, rigorously evaluated, and objectively compared. This skill provides the framework.

The key insight: Before generating new ideas, you need a baseline. The 文献库/ (literature library) serves as that baseline — it tells you what already exists so you know what's genuinely new.

Folder Structure

Initialize the IDEA folder with this structure:

IDEA/
├── 01-灵感收集/          # Idea collection & brainstorming
│   ├── 索引.md            # Index of all ideas + literature novelty map
│   ├── 想点子指南.md       # Idea generation methodology (6 methods)
│   └── 待评估点子.md       # Idea pool (unstructured brainstorming)
├── 02-评估中/            # Under evaluation
│   └── [点子名]/          # Each idea gets its own folder
│       └── [评估白皮书.md] # Evaluation white paper
├── 03-进行中/            # Actively working on
├── 04-已归档/            # Completed or abandoned
├── 05-文献库/            # Literature library — research directions extracted from papers
│   └── [论文名].md        # One file per paper (title, core idea, direction category)
└── README.md             # System overview & file responsibilities

To initialize, run:

mkdir -p IDEA/{01-灵感收集,02-评估中,03-进行中,04-已归档,05-文献库}

Then create the template files from assets/ — copy each .md file to its corresponding location.

Workflow

Phase 0: Initialize & Build Literature Baseline

This is the most critical step. Before generating any ideas, you must understand what already exists.

When the user says "initialize the IDEA folder" or "set up the IDEA system":

1. Create the folder structure (see above)
2. Create template files from assets/
3. Scan the literature library: Read every .md file in 05-文献库/
4. For each literature note, extract: title, authors, research direction, core idea
5. Classify each paper's research direction into categories (e.g., interpretability, robustness, fault diagnosis, etc.)
6. Populate 索引.md: Write all extracted research directions into the novelty verification section
7. Identify gaps: List directions that are NOT covered by any paper — these are your potential opportunity spaces

Literature note format (each file in 05-文献库/):

# [Paper Title]

- **Authors**: [Names]
- **Journal/Year**: [Venue, YYYY]
- **Direction**: [Research direction category]
- **Core Idea**: [One-paragraph summary of the method]
- **Relation to My Work**: [Why this paper matters to your research]

Even if the user hasn't populated 05-文献库/ yet, create the placeholder so they know to fill it. The literature baseline can be built incrementally — add papers as you read them.

When to update the literature library: Every time the user reads a new paper, add a note file. The index should be re-synced.

Phase 1: Generate Ideas

When the user asks to "think of new ideas" or "brainstorm research directions":

1. Read 01-灵感收集/想点子指南.md — understand the generation methods
2. Read 01-灵感收集/索引.md — understand what already exists (novelty baseline)
3. Check the literature gaps — the gaps identified in the index are priority targets
4. Apply each method systematically (see references/idea-generation-methods.md for detailed guidance)
5. Perform novelty verification via web search
6. Write results to 01-灵感收集/待评估点子.md
7. Register each new idea in 01-灵感收集/索引.md with status "待评估" — this prevents duplicate ideas during brainstorming and makes the index the single source of truth for all ideas

Phase 2: Evaluate an Idea

When the user asks to "evaluate" or "assess" an idea:

1. Read 01-灵感收集/想点子指南.md for methodology context
2. Conduct comprehensive literature research:
   - Web search with multiple keyword combinations
   - Search the 05-文献库/ literature library
   - Check user's knowledge base (if configured)
   - Search for direct competitors, adjacent work, and related methods
3. Write evaluation white paper using template in references/evaluation-template.md
4. Create idea folder in 02-评估中/[idea-name]/
5. Update 01-灵感收集/索引.md — change status from "待评估" to "评估中", fill in evaluation details (score, verdict)
6. Update 01-灵感收集/待评估点子.md to mark the idea as evaluated

Phase 3: Novelty Verification

Before claiming an idea is novel, verify through multiple channels:

1. 05-文献库/ — Check if any literature note already covers this direction
2. Web search — Use at least 3-5 different keyword combinations (English + Chinese)
3. Survey papers — Check if comprehensive surveys mention the direction
4. Knowledge base — If user has an external knowledge base (e.g., IMA, Zotero), search it
5. Claim carefully — Distinguish "no work exists" from "no exact match but related work exists"

Phase 4: Sync Literature → Index

When the user adds new papers to 05-文献库/:

1. Read each new literature note
2. Extract research direction and core idea
3. Update the novelty verification section in 索引.md
4. Re-check: do any new papers overlap with existing ideas in 索引.md (check against all statuses)?
5. If overlap found, update the affected idea's entry in 索引.md with a note about the collision

Phase 5: Transition Idea Status

When the user says "move [idea] to [new status]" or "I'm starting/abandoning this idea":

1. Identify current idea entry in 索引.md
2. Update the status:
   - 进行中 → moving from 02-评估中/ (evaluation folder) to 03-进行中/ (active)
   - 已归档 → moving from 03-进行中/ or 02-评估中/ to 04-已归档/
3. Move the idea folder to the corresponding directory
4. Add a brief summary note (e.g., "reason for archiving" if applicable)
5. Update 索引.md with the new status

Phase 6: Refine & Update

When the user says "fix" or "update" an evaluation:

1. Identify specific problems (literature gaps, wrong novelty claim, missing references)
2. Conduct targeted additional research
3. Revise the white paper, noting what changed
4. Update version number

File Responsibilities

Data Flow

文献库/ (literature library)
        ↓ scan & classify on init / sync
索引.md (novelty baseline + gaps + idea register)
        ↓ informs
想点子指南.md (methodology)
        ↓ apply methods
待评估点子.md (raw pool) ──→ 索引.md (register as pending)
        ↓ research & evaluate
02-评估中/[idea-name]/ (white paper)
        ↓ update status
索引.md (待评估 → 评估中, fill score)

Important Rules

1. Literature first, ideas second — Always build the novelty baseline before generating ideas
2. One evaluated idea = one folder — Each evaluated idea gets its own folder and white paper in 02-评估中/
3. One paper = one file — Each paper in 05-文献库/ gets its own note
4. Novelty requires proof — Never claim "this is novel" without searching at least 3 keyword combinations
5. Score honestly — A score of 3.0 means "adequate", not "good". Don't inflate
6. Index is the single source of truth — Every idea (pending or evaluated) and every literature note must be registered in 索引.md. Ideas enter the index at generation time (status: 待评估), not just after evaluation
7. Idea lifecycle — Ideas flow through statuses: 待评估 → 评估中 → 进行中 → 已归档. Update 索引.md at every transition
8. Track competitors — If related work exists, document it explicitly. Don't pretend it doesn't
9. Keep literature notes concise — One file per paper, focus on direction + core idea (not full summary)

Evaluation White Paper Structure

Each evaluation white paper should contain (see references/evaluation-template.md for full template):

1. Executive Summary — Score, verdict, key finding
2. Idea Definition — What exactly is the idea
3. Research Status — Existing work, competitors, adjacent work
4. Novelty Verification — What's been checked, what's truly new
5. Technical Feasibility — Concrete technical paths
6. Application Scenarios — Where would this be useful
7. Risk Assessment — What could go wrong
8. Literature Support — Key references
9. Horizontal Comparison — How it compares to other evaluated ideas
10. Conclusion & Recommendation — Proceed or not

Scoring System

Rate each idea on 6 dimensions (1-5 scale, 0.5 increments):

Total = sum of all 6 dimensions (max 30).