The Numbers
1,052 GitHub contributions in all of 2024. 8,563 in the last year.
That 8x jump wasn’t because I started working longer hours. It’s because I started working with Claude Code . I started experimenting in June 2025, but the real inflection point was November, when I consolidated everything into multiple concurrent sessions, automated repetitive workflows, and stopped re-explaining the same conventions every time.
Over the past 6 months, I’ve sent 14,500+ prompts across 92 projects. What used to take days now takes hours: grant proposals, manuscript preparation, Brain Imaging Data Structure (BIDS) data conversion, experiment design, CI/CD pipelines, security audits, literature reviews, and more.
What Changed
The shift wasn’t just about typing less code. It was about encoding domain knowledge into reusable workflows. Every time I taught Claude Code how to do something, whether that was formatting an NIH R01 proposal or converting EMG data to BIDS, I captured that knowledge as a plugin so I’d never have to explain it again.
These plugins are now open source as the Research Skills marketplace for Claude Code, covering 6 domains:
- Literature search and review – search across Semantic Scholar, OpenAlex, and PubMed; synthesize findings into cohesive reviews with thematic organization and gap analysis
- Grant proposal writing and review – NIH/NSF proposal drafting with mechanism-specific templates (R01, R21, K99, CAREER), self-review with simulated study section scoring, and figure quality assurance
- Manuscript preparation and peer review – IMRAD structure guidance, journal-specific formatting (IEEE, Nature, PNAS, Elsevier), LaTeX/BibTeX management, and peer review calibration
- Scientific figure creation – publication-quality figures using matplotlib, seaborn, plotly, and react-pdf composition with colorblind-safe palettes and journal dimension requirements
- Project lifecycle management – project initialization, epic/sprint workflows with git worktrees, CI/CD scaffolding, Docker packaging, security auditing, and document processing
- Neuroinformatics – BIDS conversion/validation for EEG and EMG, Hierarchical Event Descriptors (HED) annotation, and PsychoPy experiment design with Lab Streaming Layer (LSL) integration
Agentic Research Is Coming
This isn’t just about my workflow. The shift toward agentic development in research software is already underway.
EEGLAB may be the first major neuroscience software package to include a CLAUDE.md at its root, making it natively ready for AI coding agents. This means Claude Code (and similar tools) can understand the project’s conventions, architecture, and coding standards from the moment you open it.
When your codebase is agent-ready, the AI doesn’t just autocomplete lines of code. It understands your project structure, follows your conventions, runs your tests, creates pull requests, and reviews its own work. This is a fundamentally different way of building research software.
10-Week Course: AI-Powered Research Workflows
I’m putting together a live course to walk through these workflows for the research community. The format is 30-45 minute live streams with 15 minutes of Q&A, one session per week.
Who This Is For
Researchers, graduate students, postdocs, and lab managers who want to integrate AI coding agents into their academic workflows. No prior coding experience required for the first few weeks; we build up from foundations.
Syllabus
Week 1: Foundations – Git, GitHub, and the Command Line Why version control matters for research. Setting up Git, creating a GitHub account, basic commands (clone, commit, push, pull). Branching and pull requests. How this connects to everything else in the course.
Week 2: Setting Up Claude Code for Research Installing Claude Code, configuring your first project, understanding CLAUDE.md. How to write effective prompts for code generation. Setting up your development environment (UV for Python, Bun for JavaScript). Your first AI-assisted coding session.
Week 3: Project Management with AI Initializing projects with templates and standards. Epic/sprint workflows with git worktrees for multi-phase features. GitHub Issues, sub-issues, and PR-driven development. How to break complex research software into manageable phases.
Week 4: CI/CD and Code Quality Setting up GitHub Actions for automated testing and linting. Pre-commit hooks with Ruff (Python) and Biome (TypeScript). Coverage reporting, release automation, and Docker packaging. Security auditing and dependency management.
Week 5: Literature Search and Review Using opencite for academic literature search across Semantic Scholar, OpenAlex, and PubMed. Citation graph exploration, PDF retrieval, and BibTeX management. Writing literature reviews with AI: thematic organization, citation weaving, gap analysis, and avoiding common pitfalls.
Week 6: Grant Proposal Writing Drafting NIH and NSF proposals with mechanism-specific guidance (R01, R21, K99, CAREER). Specific Aims pages, Research Strategy sections, budget justification. Self-review using simulated study section scoring. Figure quality assurance for grant submissions.
Week 7: Manuscript Preparation and Peer Review Academic writing with AI: IMRAD structure, section-by-section guidance, tense and voice conventions. Journal-specific formatting (IEEE, Nature, PNAS, Elsevier). LaTeX and BibTeX management. Submission checklists and cover letters. Using AI for peer review calibration.
Week 8: Scientific Figures Creating publication-quality figures programmatically. Icon generation, data plots (matplotlib, seaborn, plotly), and multi-panel composition with react-pdf. Colorblind-safe palettes, journal dimension requirements, and the visual QA feedback loop.
Week 9: Neuroinformatics – Data Standards and Experiments Converting EEG/EMG data to BIDS format. JSON sidecars, events.tsv, channels.tsv, and validation. Designing PsychoPy experiments with proper timing, event markers, and Lab Streaming Layer (LSL) integration. BIDS-compatible output from experiments.
Week 10: Building Your Own Plugins and Workflows Creating custom Claude Code plugins for your lab or domain. Skills, agents, commands, and hooks. Encoding domain knowledge so your AI assistant remembers your conventions. Sharing plugins with your research community.
Get Involved
If this sounds useful, I’d love to hear from you. Reach out on LinkedIn or send me a message. If there’s enough interest, I’ll finalize the schedule and streaming platform.
The plugins are free and open source: github.com/neuromechanist/research-skills
Install in Claude Code:
claude plugin marketplace add neuromechanist/research-skills
for p in project grant manuscript opencite scientific-figures neuroinformatics; do
claude plugin install "$p@research-skills"
done