Claude Code plugin for research workflows: read arxiv papers, generate reports from experiment results, and submit them to a centralized dashboard.

In Claude Code, run:

/plugin marketplace add kdkyum/research-tools
/plugin install research-tools@kdkyum-research-tools

/plugin marketplace update kdkyum-research-tools

submit-report — Create ~/.dashboard.env:

DASHBOARD_URL="https://<your-dashboard-ip>:3000"
DASHBOARD_API_KEY="<your-api-key>"
DASHBOARD_USER="<your-cluster-name>"   # optional, defaults to $(hostname)

Ask the dashboard admin for the API key. Set DASHBOARD_USER if you run on SLURM compute nodes (where hostname changes per job) to keep a consistent identity.

research-report — Needs matplotlib in a project venv:

uv venv .venv
uv pip install -p .venv markdown matplotlib

The skills expect this layout (created automatically when used):

<project>/
├── research_notes/          # Self-contained report folder
│   ├── *.md                 # Markdown reports (YYYY-MM-DD-HHMMSS_<title>.md)
│   └── attachements/        # Figures, generated scripts (.png, .pdf, .py)
└── knowledge/               # Arxiv paper summaries

Auto-triggers on: "write a report", "summarize results", "document the experiment", "create research notes", "analyze these results".

Generates a structured markdown report from any experiment artifacts (JSON, CSV, Jupyter notebooks, figures, logs). Reports are saved to research_notes/YYYY-MM-DD-HHMMSS_<title>.md with figures in research_notes/attachements/.

Auto-triggers on: arxiv URLs, "read this paper", "summarize this arxiv paper".

Downloads the TeX source of an arxiv paper, reads it fully, and produces a project-contextualized summary at ./knowledge/summary_{tag}.md. The summary connects the paper's ideas to the current codebase — what techniques apply, what experiments to try, what code would change.

Paper sources are cached at ~/.cache/arxiv-papers/knowledge/{arxiv_id}/ so re-reading is instant.

Auto-triggers on: "submit report", "upload to dashboard", "push this report", "resubmit", "sync report to dashboard".

Submits a research report (and associated figures from research_notes/attachements/) to the centralized Research Dashboard. Supports:

• Auto-detection of the latest report in research_notes/
• Automatic inference of project name and tags from report content
• Versioned updates (--update) for re-submitting modified reports
• Environment and git metadata collection

Requires DASHBOARD_URL and DASHBOARD_API_KEY in ~/.dashboard.env.