Research Innovation Explorer

A host-neutral, search-first skill for literature-grounded idea discovery, theory framing, and polished Markdown reporting.

中文文档

Why This Exists

Most research-idea workflows fail in one of three ways:

• they rely on vague intuition instead of systematic search
• they generate combinations but cannot explain why the combination matters
• they stop at analysis and never produce a clean, evidence-backed report

research-innovation-explorer is built to close those gaps with one coherent workflow:

1. Search broadly and repeatedly.
2. Decompose papers into reusable capabilities.
3. Generate and score candidate combinations.
4. Frame the strongest hypothesis honestly.
5. Finish with a polished Markdown report that includes citations, analysis basis, and visual summaries.

Core Methodology

This skill is built around one explicit research-production loop:

1. Collect roughly 40 recent, top-tier, oral-level, open-source, industry-recognized papers.
2. Build a 40 x 40 directional combination matrix over those papers.
3. Remove the diagonal self-pairs and keep the remaining 40 x 39 = 1560 possible A + B combinations.
4. Run fast logical checks, lightweight experiments, and targeted search passes on those 1560 candidates.
5. Use the evidence to filter the space down to roughly 15 ideas that are actually defensible and runnable.

This is the operational core of the workflow, not a side note. The point is not to wait for a single flash of inspiration. The point is to search comprehensively, force structured combination, validate aggressively, and only then keep the few ideas that survive contact with evidence.

What You Get

Workflow

flowchart LR
    A[Search Pass] --> B[Paper Pool]
    B --> C[Capability Decomposition]
    C --> D[Idea Matrix]
    D --> E[Shortlist]
    E --> F[Theory Framing]
    F --> G[Experiment Plan]
    G --> H[Markdown Report]

Design Principles

1. Search First

The skill assumes that current literature claims should not come from memory alone when search is available.

2. Honest Framing

It supports strong abstraction and theory writing, but only when the assumptions, limiting cases, and failure boundaries are real.

3. Evidence-Carrying Reports

The final Markdown output is not a pretty wrapper around intuition. It must include:

• citations
• explicit analysis basis
• candidate comparison
• visual summaries

4. Host Neutrality

The workflow is portable across different agent hosts and even manual use. The repo does not depend on one specific runtime.

Quick Start

1. Prepare the search pack

python scripts/build_search_queries.py \
  --topic "long-context reasoning" \
  --keywords "memory routing, verifier head, benchmark"

2. Build the paper pool

Start from:

• assets/templates/search-log.csv
• assets/templates/paper-pool.csv

3. Generate the idea matrix

python scripts/build_idea_matrix.py \
  assets/templates/paper-pool.csv \
  --output work/idea-matrix.csv

4. Generate the Markdown report

Generate static figures first when the final research output should include academic paper-style data visuals:

python scripts/build_research_figures.py \
  --paper-pool assets/templates/paper-pool.csv \
  --idea-matrix work/idea-matrix.csv \
  --output-dir work/figures \
  --topic "Long-Context Reasoning" \
  --prefix long_context

python scripts/build_markdown_report.py \
  --topic "Long-Context Reasoning" \
  --paper-pool assets/templates/paper-pool.csv \
  --idea-matrix work/idea-matrix.csv \
  --search-log assets/templates/search-log.csv \
  --figure-dir work/figures \
  --figure-prefix long_context \
  --output work/report.md

Output Style

The reporting layer is intentionally designed for GitHub-native reading:

• Mermaid flowcharts for process explanation
• static PNG heatmaps for matrix snapshots and worked examples
• Mermaid pie charts for quick distribution views
• Markdown evidence tables for claim tracing
• compact narrative sections for executive summary and detailed analysis

This makes the output readable both as a working note and as a shareable artifact.

Example Output

Exploring LLM Training Directions

This worked example uses frontier large language model training research as the target domain. It starts from a search-backed pool of roughly 40 recent, top-tier, oral-level, open-source, industry-recognized papers, builds a 40 x 40 directional matrix, removes the diagonal, and then searches, inspects, and lightly validates the remaining 1560 A + B candidates until only a small shortlist remains.

At the survey level, the workflow turns the literature into a readable interaction matrix instead of a prose dump:

At the decision level, the workflow collapses the shortlist into a result matrix that exposes novelty space, implementation feasibility, evaluation clarity, open-source readiness, and narrative sharpness:

What this example demonstrates:

• search is used during collection and during analysis, not only at the beginning
• the 40 x 40 -> 1560 -> ~15 narrowing process is explicit and inspectable
• GitHub README pages and Markdown reports can show the screening logic with direct images, without depending on host-side math rendering

The bundled example images live in assets/examples/llm-training/ and can be regenerated with scripts/build_llm_training_example_figures.py.

Repository Layout

.
├── SKILL.md
├── README.md
├── README.zh-CN.md
├── agents/
│   └── openai.yaml
├── assets/
│   ├── examples/
│   │   └── llm-training/
│   └── templates/
├── references/
└── scripts/
    ├── build_idea_matrix.py
    ├── build_llm_training_example_figures.py
    ├── build_markdown_report.py
    ├── build_research_figures.py
    └── build_search_queries.py

Recommended Use Cases

• discovering incremental but defensible research ideas
• mapping literature around a topic before starting implementation
• checking whether an A+B combination already exists in prior work or code
• producing a strong Markdown research memo with citations and visual summaries
• training literature review, abstraction, evaluation design, and research writing habits

Documentation

• Core workflow: SKILL.md
• Search protocol: references/search-playbook.md
• Theory framing: references/framing-and-theory.md
• Reporting rules: references/reporting-and-visualization.md
• Report template: assets/templates/analysis-report-template.md

Notes

• If your host cannot render Mermaid, keep the Markdown tables and replace Mermaid blocks with static images or plain-text summaries.
• If your host has no search capability, use the workflow manually and explicitly downgrade confidence in current-literature claims.

Community

For broader discussion around tools, workflows, and AI-native building, visit linux.do.

License

This repository is released under the MIT License.