research-brainstorm-from-kb

name	research-brainstorm-from-kb
follows	rf-obsidian-markdown
description	Explores and structures research ideas into `obsidian-vault/ideas/` notes using the local paper knowledge base and frontier techniques. Use when the user provides research questions or ideas and wants decomposed candidate directions, related-work analysis based primarily on `obsidian-vault/analysis`, and cross-domain support from image/video generation, MLLM, Agent, or RL, automatically saved as dated idea notes under `obsidian-vault/ideas/`.

Research Idea Brainstorming

1. Use cases (When to use)

Use this skill in the following scenarios:

• Research question / idea input: the user describes a research question, direction, or fragmented ideas in natural language and wants systematic brainstorming plus lightweight synthesis.
• Need local knowledge-base grounding: rely primarily on retrieval from obsidian-vault/analysis, and use generated obsidian-vault/index indexes/navigation pages when they help produce "related-work support + research opportunity".
• Need reusable idea notes: automatically write outputs as structured Markdown under local obsidian-vault/ideas/ for later writing, experiments, and project management.

2. Dependencies and paths (Dependencies & paths)

This skill depends on the local paper knowledge base and the papers-query-knowledge-base skill:

• Paper index and analysis: see papers-query-knowledge-base skill (obsidian-vault/index/ + obsidian-vault/analysis/)
• Primary paper retrieval and analysis: see papers-query-knowledge-base skill (obsidian-vault/analysis/ for evidence, generated obsidian-vault/index/ for fast filtering/navigation when useful)
• Idea note directory: obsidian-vault/ideas/

Path convention:

• If the current workspace is the repository containing these folders, prefer relative paths.
• If the workspace does not include this KB, use the correct absolute path on your machine.

3. File naming and storage rules

When generating brainstorming results, always write/update Markdown files in obsidian-vault/ideas/ using this naming rule:

• Filename pattern: YYYY-MM-DD_<core-slug>.md
  • YYYY-MM-DD: user local date (for example 2025-03-09)
  • <core-slug>: topic phrase compressed into English words or pinyin, lowercase with - separators (for example motion-llm-ideas, reactive-agent-motion)
• Example: 2025-03-09_motion-llm-ideas.md

Behavior convention:

• If a file with the same date and core slug already exists, append a new brainstorming subsection instead of creating a new file.
• If the user explicitly specifies a target file in obsidian-vault/ideas, follow the user-specified filename.

4. Brainstorming process and output structure

Overall goal: starting from the user's raw idea, combine the local paper KB and frontier techniques to complete one end-to-end brainstorming-and-synthesis cycle and leave a note structure directly usable for writing/topic selection.

4.1 Idea decomposition and association

Steps:

1. Refine the problem core: restate the user problem in 1-3 sentences and identify task/scenario/key bottleneck.
2. Multi-dimensional decomposition: split into sub-problems by task, data (modality/source), model (architecture), constraints (physical/semantic/interaction), and evaluation.
3. Lateral association: map to related themes in the local KB (for example motion generation, HSI, HOI, Agent integration), and identify possible transfer directions or variants.

In generated notes, use this section:

• Section title: ## 1. Idea decomposition and association

4.2 Real scenarios and pain points

Steps:

1. Lock real scenarios: based on assumptions or user description, define representative application scenarios (for example VR interaction, rehabilitation training, animation production, robot manipulation).
2. Enumerate needs and pain points:
   • How current practice/systems solve this.
   • Which steps still rely heavily on manual work, low efficiency, or poor UX.
   • Hidden requirements such as safety, fairness, robustness.
3. Map pain points back to the idea: show where the original idea directly alleviates pain or amplifies value.

In generated notes:

• Section title: ## 2. Real scenarios and pain points

4.3 Related-work support and research opportunities (via papers-query-knowledge-base)

Always use the local KB through papers-query-knowledge-base:

1. Locate related tasks/techniques:
   • Prioritize title, task path, tags, venue, year, core_operator, and primary_logic in obsidian-vault/analysis/.
   • If fast filtering or overview/statistics/Obsidian navigation aid is needed, reference obsidian-vault/index/index.jsonl, by_topic/, by_method/, by_dataset/, and by_venue_year/.
   • For venue/year filtering, prefer the structured venue, year, and venue_year fields in index.jsonl; the Obsidian navigation layer uses merged labels such as ICLR_2026 under by_venue_year/.
2. Select representative papers:
   • Choose 3-8 papers highly related to the idea.
   • For each paper, read core_operator and primary_logic from the obsidian-vault/analysis note frontmatter, then read Summary and Key Performance from the note body TL;DR section.
   • Summarize in 1-2 sentences what it does and how it relates to the current idea.
3. Summarize support and gaps:
   • Support: where these works validate the idea's feasibility or value.
   • Gap/opportunity: where research space remains (scenario, data, constraints, evaluation, degree of systemization).

In generated notes:

• Section title: ## 3. Related-work support and research opportunities
• Suggested content layout:
  • "Related-work overview" subsection listing papers by task/technique cluster with one-line summaries.
  • "Support points" subsection listing strong evidence for the idea.
  • "Research opportunities" subsection listing potentially top-tier innovation entry points.

4.4 Frontier cross-domain techniques and validation (Image / Video / MLLM / Agent / RL)

Goal: connect the current idea to a broader frontier ecosystem, propose viable cross-domain research paths, and provide links for further reading.

Steps:

1. Identify suitable frontier directions:
   • image/video generation (diffusion, video diffusion, 3D/4D representation)
   • multimodal LLM (MLLM), multimodal Agent
   • reinforcement learning (RL), reward/preference-guided generation
   • popular technical domains or transferable core ideas
   • other high-fit techniques (for example 3DGS, Neural ODE, Conformal prediction)
2. Retrieve and filter:
   • prioritize relevant existing local KB analysis notes if available.
   • for newer uncovered methods/applications, use web search for representative papers, technical blogs, or official docs.
3. Provide integration and validation plans:
   • explain how each technique plugs into the current idea pipeline (as backbone, module, evaluation tool, or Agent sub-skill).
   • provide simple validation plans or prototype experiment concepts.

In generated notes:

• Section title: ## 4. Frontier cross-domain techniques and validation ideas
• Include a small table/list with this format:
  • technique name / direction
  • brief description (1-2 sentences)
  • related links (paper/project/blog), using Markdown links such as [Paper Title](https://...) (no bare URLs)

4.5 Summary and next actions

Finally provide a convergent summary:

1. Summarize the core idea and its research position: use 3-5 sentences to explain what problem is solved, key ideas used, and where novelty sits relative to existing work.
2. List executable next steps:
   • data/scenario: what data or scenarios need to be built/organized?
   • baseline/experiments: which existing methods can bootstrap baselines quickly?
   • metrics/evaluation: how to quantify the idea advantage?
3. Optional: venue recommendation: if appropriate, suggest 1-2 suitable venues and rationale.

In generated notes:

• Section title: ## 5. Summary and next steps

5. Recommended note template

When generating obsidian-vault/ideas notes, you can use this template (slight adjustments allowed by context):

---
hypothesis: "{{One-sentence research hypothesis}}"
status: brainstorm
source_papers: []
created: {{ISO_DATETIME_NOW}}
updated: {{ISO_DATETIME_NOW}}
---

# {{YYYY-MM-DD}} {{Brief core problem description}}

> Systematic retrieval and brainstorming primarily grounded in `obsidian-vault/analysis`, using generated `obsidian-vault/index` index/navigation support when helpful, with frontier cross-domain support from image/video/MLLM/Agent/RL.

---

## 1. Idea decomposition and association
- Problem restatement:
- Key elements:
- Multi-dimensional decomposition:

## 2. Real scenarios and pain points
- Typical scenarios:
- Core needs:
- Existing solutions and pain points:

## 3. Related-work support and research opportunities
### 3.1 Related-work overview
- Paper A: one-line summary + relation to this idea. Put the full local wikilink in `source_papers` or in a nearby list outside tables.
- Paper B: one-line summary + relation to this idea. External papers/projects use Markdown links in the body, not `source_papers`.

### 3.2 Support points
- ...

### 3.3 Research opportunities
- ...

## 4. Frontier cross-domain techniques and validation ideas
- Technique direction A: brief description + [link](https://...)
- Technique direction B: brief description + [link](https://...)

## 5. Summary and next steps
- Core idea summary:
- Near-term executable steps:
- Potential target venue:

6. Implementation notes

• Always prioritize papers-query-knowledge-base to locate existing local analyses, then supplement with frontier web search.
• Use current index contract: index.jsonl is the fast filter layer; navigation pages are by_topic/, by_method/, by_dataset/, and by_venue_year/. Do not reference legacy by_venue/ or by_year/.
• Use current analysis schema: frontmatter provides title, venue, year, tags, aliases, pdf_ref, core_operator, primary_logic, and optional claims; TL;DR summary and key performance are in the body.
• Use current idea schema: idea notes should include hypothesis, status, and source_papers frontmatter. source_papers is only for local KB analysis-note wikilinks actually used as support; external web papers/blogs stay as Markdown links in the body.
• Ensure structured Markdown output following the section structure above for later retrieval/recomposition.
• When citing external resources, always use Markdown links and avoid bare URLs.
• If the user already opened a specific obsidian-vault/ideas file and explicitly asks to brainstorm there, append the structured subsection in that file instead of creating a new one.

7. Boundaries

• This skill can refine framing and next actions, but detailed scope cuts, prioritized hypotheses, and MVP planning belong to idea-focus-coach.
• This skill is for candidate-direction generation, not reviewer-style verdicts or acceptance-risk scoring. Use reviewer-stress-test for that.