Skip to content
Back to search

research-innovation-explorer

The automated approach leverages the cross-combination of high-quality papers from top conferences to uncover feasible research and innovation ideas. Through multi-level verification and convergence screening, it identifies research schemes that are feasible and have in-depth value.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/foryourhealth111-pixel--research-innovation-explorer/badge.svg)](/listings/foryourhealth111-pixel--research-innovation-explorer)

HTML

<a href="/listings/foryourhealth111-pixel--research-innovation-explorer"><img src="/api/listings/foryourhealth111-pixel--research-innovation-explorer/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

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. Review candidates and optionally generate a report

After generating the matrix, copy `assets/templates/candidate-review.yaml` for candidates in the review queue. Read `references/post-matrix-review.md` and record source-linked facts, inferences, status, confidence, and next checks.

The report script remains available as a matrix-overview scaffold. Generate static figures and a Markdown overview only when they help the current review:

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

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 75 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. ---
  2. name: research-innovation-explorer
  3. description: Build literature-grounded research-question candidate landscapes by collecting papers, generating A+B matrices, and dynamically reviewing combinations with traceable evidence, uncertainty, and next checks. Use when an AI agent needs to explore a field, screen research questions, compare paper combinations, inspect prior art, or prepare a provisional shortlist for researcher review.
  4. ---
  5. # Research Innovation Explorer
  6. ## Purpose
  7. Help a researcher decide which literature-grounded questions deserve more attention. Search broadly, structure a paper pool, generate an A+B matrix, and review selected candidates against source evidence.
  8. Stop the default workflow at a provisional candidate landscape. Do not present matrix rankings as novelty, feasibility, publishability, or expected research success. Expand a researcher-selected candidate into theory framing, an experiment plan, or a publication-oriented report only when requested.
  9. Use this skill as a host-neutral contract. Adapt search and browsing actions to the tools available in the current environment.
  10. ## Default Workflow
  11. 1. Clarify the topic, resource constraints, available data or code, and the desired breadth of the candidate landscape.
  12. 2. Read `references/search-playbook.md`, create a working `search-log.csv`, and generate a starter query pack with `scripts/build_search_queries.py`.
  13. 3. Build a 20-50 paper pool and normalize each paper into tasks, mechanisms, strengths, weaknesses, benchmarks, and implementation signals. Read `references/workflow.md` for intake rules.
  14. 4. Run `scripts/build_idea_matrix.py` to generate `idea-matrix.csv`. The script rejects duplicate paper identifiers and orders papers by identifier so each pair has stable A/B roles.
  15. 5. Treat matrix scores only as queue-priority signals. Build the review queue from the ten highest-ranked unique pairs, up to five coverage-increasing pairs, and every researcher-requested pair.
  16. 6. Read `references/post-matrix-review.md` and `references/scoring-rubric.md`. Copy `assets/templates/candidate-review.yaml` for each candidate under review.
  17. 7. Complete both entries under `direction_checks`. Select a direction only after both `A -> B` and `B -> A` have been assessed.
  18. 8. In each review round, identify the single uncertainty most likely to change the recommendation. Perform one focused action, then update facts, inferences, the relevant direction check, decision-linked inference identifiers, status, confidence, and the next check.
  19. 9. Validate each populated review with `scripts/validate_candidate_review.py` before including it in the candidate landscape.
  20. 10. Stop when further searching mainly repeats known information, a status is adequately supported, or the next decision requires researcher input. Preserve unresolved uncertainty in the record.
  21. 11. Produce a candidate landscape using `references/reporting-and-visualization.md` and `assets/templates/analysis-report-template.md`.
  22. ## Candidate Review Rules
  23. - Verify source facts before interpreting a pairing.
  24. - Preserve the canonical `paper_a_id`, `paper_b_id`, and `candidate_id` defined by the matrix order.
  25. - Record observed facts separately from agent inferences.
  26. - Give every observed fact a stable source URL and a section, page, figure, table, or repository location.
  27. - Link every inference to the fact identifiers that support it.
  28. - Link each non-unknown direction assessment, selected direction, and decisive research status to inference identifiers.
  29. - Use `unknown` only for direction or dimension fields. Use `needs_check` or `conflicting` for research status when evidence warrants them.
  30. - Keep matrix score and qualitative review judgment separate. Do not calculate a second aggregate score.
  31. - Reopen any research judgment when new evidence changes the basis.
  32. - Stop review on broken input data and resume after the data is repaired.
  33. ## Default Deliverables
  34. - `search-log.csv`
  35. - `paper-pool.csv`
  36. - `idea-matrix.csv`
  37. - one `candidate-review.yaml` per reviewed candidate
  38. - one candidate-landscape Markdown document covering promising, unresolved, parked, weak, and excluded candidates
  39. - optional screening figures when visual comparison is useful
  40. ## Optional Expansion
  41. After the researcher selects a candidate:
  42. - read `references/framing-and-theory.md` for a framing note
  43. - read `references/experiment-plan.md` and use `assets/templates/experiment-plan.md` for a validation plan
  44. - use `scripts/build_research_figures.py` for screening visualizations
  45. - use `scripts/build_markdown_report.py` only as a matrix-overview scaffold, then add the candidate-review evidence manually
  46. Keep all claims proportional to the available evidence. Read `references/ethics-boundaries.md` whenever wording about novelty, theory, or expected results becomes stronger than the sources support.
  47. ## Resources
  48. - `references/workflow.md`: paper intake, matrix generation, queue construction, and default outputs
  49. - `references/search-playbook.md`: search objectives, source selection, logging, and stopping rules
  50. - `references/post-matrix-review.md`: dynamic review loop, evidence records, statuses, and failure handling
  51. - `references/scoring-rubric.md`: qualitative dimensions and state assignment
  52. - `references/reporting-and-visualization.md`: candidate-landscape reporting rules
  53. - `references/framing-and-theory.md`: optional framing guidance for selected candidates
  54. - `references/experiment-plan.md`: optional experiment planning guidance
  55. - `references/ethics-boundaries.md`: claim and evidence boundaries
  56. - `assets/templates/candidate-review.yaml`: stable review record interface
  57. - `assets/templates/analysis-report-template.md`: researcher-facing candidate landscape
  58. - `scripts/validate_candidate_review.py`: deterministic candidate identity and evidence-link validation
README.md · README · 252 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. <div align="center">
  2. # Research Innovation Explorer
  3. **This search-first workflow turns a structured paper pool and an A+B matrix into an evidence-grounded landscape of research questions, uncertainties, and next checks for researcher review.**
  4. [中文文档](./README.zh-CN.md)
  5. [![GitHub stars](https://img.shields.io/github/stars/foryourhealth111-pixel/research-innovation-explorer?style=flat-square)](https://github.com/foryourhealth111-pixel/research-innovation-explorer)
  6. [![GitHub repo size](https://img.shields.io/github/repo-size/foryourhealth111-pixel/research-innovation-explorer?style=flat-square)](https://github.com/foryourhealth111-pixel/research-innovation-explorer)
  7. ![License: MIT](https://img.shields.io/badge/license-MIT-166534?style=flat-square)
  8. ![Host-neutral](https://img.shields.io/badge/host-neutral-111827?style=flat-square)
  9. ![Search-first](https://img.shields.io/badge/search-first-0f766e?style=flat-square)
  10. ![Markdown reporting](https://img.shields.io/badge/reporting-markdown%20%2B%20visuals-7c3aed?style=flat-square)
  11. <div align="center">
  12. <img src="https://github.com/user-attachments/assets/712e718c-3521-407c-bae3-1317e5225f8d" width="40%" />
  13. <img src="https://github.com/user-attachments/assets/f6028600-eee9-4a09-b80d-f22a6b79347e" width="40%" />
  14. </div>
  15. </div>
  16. ## Why This Exists
  17. Most research-idea workflows fail in one of three ways:
  18. - they rely on vague intuition instead of systematic search
  19. - they generate combinations but cannot explain why the combination matters
  20. - they lose the evidence and uncertainty that should guide the next research decision
  21. `research-innovation-explorer` is built to close those gaps with one coherent workflow:
  22. 1. Search broadly and repeatedly.
  23. 2. Decompose papers into reusable capabilities.
  24. 3. Generate and rank candidate combinations for review.
  25. 4. Check both combination directions against source evidence.
  26. 5. Return a provisional candidate landscape with supporting evidence, uncertainty, and next checks.
  27. Theory framing, experiment planning, and publication-oriented reporting remain available as explicit follow-up layers for a researcher-selected candidate.
  28. ## Core Methodology
  29. This skill is built around one explicit research-production loop:
  30. 1. Collect roughly 40 relevant, high-quality papers with enough detail to support comparison.
  31. 2. Build a pairwise combination matrix over those papers.
  32. 3. Keep one row per unique paper pair; 40 papers produce `40 x 39 / 2 = 780` rows with the current generator.
  33. 4. Use matrix scores to build a review queue, then examine both `A -> B` and `B -> A` through focused source, prior-art, code, or benchmark checks.
  34. 5. Keep a provisional set of questions with evidence, contrary evidence, uncertainty, and recommended next actions.
  35. 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.
  36. | Stage | What to do | What comes out |
  37. | --- | --- | --- |
  38. | Paper pool | Gather around 40 relevant papers with reproducible detail | a reusable capability inventory |
  39. | Combination pass | Enumerate every unique paper pair | 780 pair rows for a 40-paper pool |
  40. | Post-matrix review | Check both directions, inspect evidence, and identify the highest-value unresolved question | candidate review records |
  41. | Provisional landscape | Group promising, unresolved, parked, weak, and excluded candidates with reasons | a researcher-ready shortlist |
  42. ## What You Get
  43. | Layer | What it does |
  44. | --- | --- |
  45. | `SKILL.md` | Defines the default exploration workflow, evidence rules, and optional expansion paths |
  46. | `scripts/build_search_queries.py` | Generates structured query packs for topic scan, novelty checks, and failure analysis |
  47. | `scripts/build_idea_matrix.py` | Builds a scored pairwise candidate matrix from the paper pool |
  48. | `scripts/build_research_figures.py` | Generates publication-style literature heatmaps, scoring heatmaps, and analysis panels from the research artifacts |
  49. | `scripts/build_markdown_report.py` | Scaffolds a Markdown matrix overview; reviewed evidence is added afterward |
  50. | `references/` | Contains the search playbook, theory framing rules, reporting rules, and ethics boundaries |
  51. | `assets/templates/` | Provides CSV, candidate-review, idea-brief, experiment-plan, and report templates |
  52. ## Workflow
  53. ```mermaid
  54. flowchart LR
  55. A[Search Pass] --> B[Paper Pool]
  56. B --> C[Capability Decomposition]
  57. C --> D[Idea Matrix]
  58. D --> E[Review Queue]
  59. E --> F[Evidence Review]
  60. F --> G[Candidate Landscape]
  61. G -. on request .-> H[Theory Framing]
  62. G -. on request .-> I[Experiment Plan]
  63. G -. on request .-> J[Extended Report]
  64. ```
  65. ## Design Principles
  66. ### 1. Search First
  67. The skill assumes that current literature claims should not come from memory alone when search is available.
  68. ### 2. Dynamic Review
  69. Each review round targets the uncertainty most likely to change the recommendation. Unknown, incomplete, and conflicting evidence remain visible in the output.
  70. ### 3. Evidence-Carrying Reports
  71. The default candidate landscape includes:
  72. - citations
  73. - observed facts and agent inferences kept separate
  74. - candidate comparison
  75. - uncertainty and next checks
  76. Matrix scores are triage signals. They do not establish novelty, feasibility, publishability, or expected research success.
  77. ### 4. Host Neutrality
  78. The workflow is portable across different agent hosts and even manual use. The repo does not depend on one specific runtime.
  79. ## Quick Start
  80. ### 1. Prepare the search pack
  81. ```bash
  82. python scripts/build_search_queries.py \
  83. --topic "long-context reasoning" \
  84. --keywords "memory routing, verifier head, benchmark"
  85. ```
  86. ### 2. Build the paper pool
  87. Start from:
  88. - `assets/templates/search-log.csv`
  89. - `assets/templates/paper-pool.csv`
  90. ### 3. Generate the idea matrix
  91. ```bash
  92. python scripts/build_idea_matrix.py \
  93. assets/templates/paper-pool.csv \
  94. --output work/idea-matrix.csv
  95. ```
  96. ### 4. Review candidates and optionally generate a report
  97. After generating the matrix, copy `assets/templates/candidate-review.yaml` for candidates in the review queue. Read `references/post-matrix-review.md` and record source-linked facts, inferences, status, confidence, and next checks.
  98. The report script remains available as a matrix-overview scaffold. Generate static figures and a Markdown overview only when they help the current review:
  99. Generate static figures first when the final research output should include academic paper-style data visuals:
  100. ```bash
  101. python scripts/build_research_figures.py \
  102. --paper-pool assets/templates/paper-pool.csv \
  103. --idea-matrix work/idea-matrix.csv \
  104. --output-dir work/figures \
  105. --topic "Long-Context Reasoning" \
  106. --prefix long_context
  107. ```
  108. ```bash
  109. python scripts/build_markdown_report.py \
  110. --topic "Long-Context Reasoning" \
  111. --paper-pool assets/templates/paper-pool.csv \
  112. --idea-matrix work/idea-matrix.csv \
  113. --search-log assets/templates/search-log.csv \
  114. --figure-dir work/figures \
  115. --figure-prefix long_context \
  116. --output work/report.md
  117. ```
  118. ## Optional Report Style
  119. The reporting layer is intentionally designed for GitHub-native reading:
  120. - Mermaid flowcharts for process explanation
  121. - static PNG heatmaps for matrix snapshots and worked examples
  122. - Mermaid pie charts for quick distribution views
  123. - Markdown evidence tables for claim tracing
  124. - compact narrative sections for executive summary and detailed analysis
  125. This makes an optional overview readable as a working note and a shareable artifact. The default deliverable remains the provisional candidate landscape.
  126. ## Example Output
  127. ### Exploring LLM Training Directions
  128. 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 papers, builds the combination matrix, and then reviews selected candidates to produce a provisional landscape with evidence and open questions.
  129. At the survey level, the workflow turns the literature into a readable interaction matrix instead of a prose dump:
  130. ![LLM training theme interaction matrix](./assets/examples/llm-training/theme_interaction_heatmap.en.png)
  131. At the decision level, the workflow uses the matrix as a screening view and records source-based questions, concerns, and next checks separately:
  132. ![LLM training shortlist evaluation heatmap](./assets/examples/llm-training/shortlist_heatmap.en.png)
  133. What this example demonstrates:
  134. - search is used during collection and during analysis, not only at the beginning
  135. - the matrix-to-review-queue process is explicit and inspectable
  136. - GitHub README pages and Markdown reports can show the screening logic with direct images, without depending on host-side math rendering
  137. The bundled example images live in [`assets/examples/llm-training/`](./assets/examples/llm-training/) and can be regenerated with [`scripts/build_llm_training_example_figures.py`](./scripts/build_llm_training_example_figures.py).
  138. ## Repository Layout
  139. ```text
  140. .
  141. ├── SKILL.md
  142. ├── README.md
  143. ├── README.zh-CN.md
  144. ├── agents/
  145. │ └── openai.yaml
  146. ├── assets/
  147. │ ├── examples/
  148. │ │ └── llm-training/
  149. │ └── templates/
  150. ├── references/
  151. └── scripts/
  152. ├── build_idea_matrix.py
  153. ├── build_llm_training_example_figures.py
  154. ├── build_markdown_report.py
  155. ├── build_research_figures.py
  156. └── build_search_queries.py
  157. ```
  158. ## Recommended Use Cases
  159. - discovering literature-grounded research questions worth further review
  160. - mapping literature around a topic before starting implementation
  161. - checking whether an A+B combination already exists in prior work or code
  162. - producing a provisional candidate landscape with citations and visual summaries
  163. - training literature review, abstraction, evaluation design, and research writing habits
  164. ## Documentation
  165. - Core workflow: [`SKILL.md`](./SKILL.md)
  166. - Search protocol: [`references/search-playbook.md`](./references/search-playbook.md)
  167. - Post-matrix review: [`references/post-matrix-review.md`](./references/post-matrix-review.md)
  168. - Candidate template: [`assets/templates/candidate-review.yaml`](./assets/templates/candidate-review.yaml)
  169. - Theory framing: [`references/framing-and-theory.md`](./references/framing-and-theory.md)
  170. - Reporting rules: [`references/reporting-and-visualization.md`](./references/reporting-and-visualization.md)
  171. - Report template: [`assets/templates/analysis-report-template.md`](./assets/templates/analysis-report-template.md)
  172. ## Notes
  173. - If your host cannot render Mermaid, keep the Markdown tables and replace Mermaid blocks with static images or plain-text summaries.
  174. - If your host has no search capability, use the workflow manually and explicitly downgrade confidence in current-literature claims.
  175. ## Community
  176. For broader discussion around tools, workflows, and AI-native building, visit [linux.do](https://linux.do/).
  177. ## License
  178. This repository is released under the [MIT License](./LICENSE).

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.