research-shelf
jasonbhorne/research-shelf
A library of AI-generated deep research reports built with Claude Code's multi-agent pipeline. 27 reports, 2 reusable skills, and the generation scripts.
Instalação
npx convez-ai jasonbhorne/research-shelf1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
The Research Shelf
A growing library of deep dives, companion guides, and curiosity-driven research, all produced using Claude Code's multi-agent research pipeline.
This repository contains 27 research reports spanning K-12 education policy, AI in education, district operations, history, and whatever else caught my attention. Each report was generated by spawning 4-5 parallel research agents that independently gather sources from different categories (government, academic, news, industry), then cross-referencing their findings, verifying URLs, and synthesizing everything into a publication-ready document.
How It Works
The pipeline is built on two Claude Code skills that orchestrate multi-agent research:
/deep-research (4 agents)
Spawns four agents, each assigned a distinct source category (government, news, academic, industry). Agents research independently, then findings are cross-referenced in a claim matrix, URLs are verified, and confidence scores are aggregated. Produces a .docx with a source reliability matrix.
/research-report (3-5 agents)
A unified pipeline that adds a case-studies agent, a fact-check pass, and optional slide/infographic generation. Supports --education mode for Tennessee-specific policy context and --light mode for faster, 3-agent runs.
Both skills are in the skills/ directory and can be dropped into any Claude Code setup.
Generation Scripts
The scripts/ directory contains the Python scripts (using python-docx) that generate the final formatted .docx files. These are bespoke per report, showing the full end-to-end pipeline from research findings to publication-ready document.
Reports
K-12 Education
| Report | Description |
|---|---|
| AI in K-12 Education: Student Outcomes | Comprehensive review of AI tools in K-12 education, including intelligent tutoring systems and adaptive learning |
| Screen Time and Academic Performance | Synthesizes 51 findings from 45 sources on how screen time impacts K-12 academic performance internationally |
| Negative Impact of Technology in K-12 | Meta-analysis of peer-reviewed studies examining how classroom technology adoption affects learning outcomes |
| Professional Learning Communities | Research on how PLCs affect student outcomes in K-12 schools |
| Missouri Through-Year Testing | Analysis of Missouri's innovative through-year assessment program and implications for other states |
| School Operations and Student Achievement | How non-instructional operations (maintenance, nursing, transportation, nutrition) affect student achievement |
AI in Education
| Report | Description |
|---|---|
| AI District Operations by Department | Practical playbook for how AI can transform seven core operational departments in a small school district |
| AI for RTI, Tutoring, and Student Outcomes | Research synthesis on AI tools for response-to-intervention and tutoring |
| AI School Operations Playbook | Evidence-based playbook for K-12 district leaders adopting AI across operations |
| AI Slop: Examples, Scale, and Impact | The scale and impact of low-quality AI-generated content proliferating across the internet |
| AI in University Settings | Synthesis of 50+ studies on how university students perceive and use generative AI |
| Education Administrators Using Claude Code | Deep research on K-12 administrator adoption of Claude Code |
| Claude Code in 60 Days | Quantitative analysis of Claude Code usage across 148 sessions in education operations workflows |
| Does Saying Please to AI Help? | Whether politeness affects AI response quality, and what those extra tokens cost at scale |
District Operations
| Report | Description |
|---|---|
| Negative Lunch Balances | How districts address school meal debt through collection, alternative meals, and universal free programs |
| School Nutrition Budget Constraints | Why school nutrition budgets constrain cafeteria menu options |
History and Culture
| Report | Description |
|---|---|
| Garfield, Arthur, and Death by Lightning | The assassination of President Garfield, the succession crisis, and the Gilded Age |
| Second Sons in Primogeniture | How primogeniture shaped career paths for younger sons in aristocratic families |
Personal Interest
| Report | Description |
|---|---|
| Thumbs Up as Passive-Aggressive | How the thumbs-up emoji became a contested symbol of passive-aggressive communication |
| Diet Drinks vs. Sugar-Sweetened Beverages | Comparing health effects on adolescent health |
| Feathered Flags and Liability | Liability risks and legal implications of feathered advertising flags |
Media Companions
| Report | Description |
|---|---|
| Severance | Apple TV+ series on work-life balance, corporate control, and identity |
| Radical Acceptance | Tara Brach's book on mindfulness and Buddhist psychology |
| Project Hail Mary | Andy Weir's novel about space survival and interstellar friendship |
| In France Profound | T.D. Allman's posthumous memoir about a French mountain town |
| Following the Sun | Margaret Bensfield Sullivan's family travel memoir |
| The Hard Thing About Hard Things | Ben Horowitz on navigating difficult business challenges |
Using the Skills
To use these research skills in your own Claude Code setup:
- Copy the skill files from
skills/into your.claude/skills/directory - Update the file paths in the skills to match your environment
- Run
/deep-research <your topic>or/research-report <your topic>
The skills require Claude Code with agent capabilities (Team, Task, and Agent tools). They work best with models that support parallel tool use.
About
Built by Jason Horne, Assistant Director of Schools for Administration at Greeneville City Schools and adjunct professor at East Tennessee State University. All research produced using Claude Code's multi-agent research pipeline.
License
Reports are provided as-is for educational and informational purposes. The research skills and scripts are open source under the MIT License.