ai-science-toolkit plugin
26 skills and four reviewer agents for science in Claude Code: figure and statistics review, literature search, work handoff.
- 83 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The install script writes to `~/.claude/` and can edit `settings.json` with a boot hook
- Some skills send data to Zotero, Notion and Slack once you configure them
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/ai-science-toolkitDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The install script writes to `~/.claude/` and can edit `settings.json` with a boot hook
- Some skills send data to Zotero, Notion and Slack once you configure them
This is third-party code. Review the repository files before installing.
What it does
ai-science-toolkit combines skills, reviewer agents and utilities for scientific work with Claude Code. Skills are slash commands: `/figure-review` checks a figure for publication readiness (colormaps, uncertainty, axes, caption), `/lit-review` searches and synthesizes literature from Zotero, arXiv, bioRxiv and other sources, `/reviewer-2` takes a strict reviewer stance and stress-tests a claim, and `/disclose` and `/disclose-project` produce a line and a final statement about AI involvement. Among the others are `/ai-review` (a repo review by parallel fan-out), `/commit-batch` (splitting a working tree into logical commits), `/handoff` and `/pickup` (passing work to the next session through `.ai/HANDOFF.md`), `/recover-sessions`, `/repo-init`, `/retro` and `/unstale`. Four reviewer agents check climate attribution claims, statistics, meteorological conclusions and science communication, individually or as a panel. `/pathfinder` shows a map of the whole set.
Who it is for. Scientists and computational engineers who write code and papers with Claude Code and want built-in strict review and work handoff.
Good fit when
- A figure for a paper needs a check of colormap, uncertainty, axes and caption before submission
- A claim or manuscript section should pass through a critical reviewer: baseline, counterfactual
- A session is ending and project state must be saved for the next agent in `.ai/HANDOFF.md`
- Before publishing you need a statement of AI involvement by phase of work
- A set of edits in the working tree must be split into single-concern commits
Not a fit when
- Work happens in a different agent: the skills and agents install as Claude Code commands
- You need a ready scientific model or calculation package: the set helps review and structure work rather than compute
- Zotero, Notion or Slack must not be connected: some skills use these services and need setup
Example request
/figure-review Check figure 3 in this paper: the colormap, how uncertainty is shown, the axes and the caption, and note what looks deliberate.Limitations
Built for Claude Code. Most skills need no setup, while writing to Zotero, the Notion work journal and session naming need configuration from `docs/configuration.md`. `sync.sh push` writes skills to `~/.claude/skills/` and agents to `~/.claude/agents/` and edits `~/.claude/settings.json` when `tab-setup` is installed. For Windows the author recommends the plugin route with no shell. The "spectrum of AI involvement" estimate in `/disclose` is approximate and a person confirms it.
How to disable. Run `/plugin uninstall ai-science-toolkit@ai-science-toolkit`. If you installed through `sync.sh`, delete the relevant folders from `~/.claude/skills/` and `~/.claude/agents/` and remove the boot hook entry from `~/.claude/settings.json`.
FAQ
How do I install the set?
With `/plugin marketplace add dgilford/ai-science-toolkit` and `/plugin install ai-science-toolkit@ai-science-toolkit`. All skills and agents install. To pick individual pieces use `bash scripts/sync.sh push figure-review reviewer-2`.
Which reviewer agents are there?
Four: `attribution-reviewer` for climate attribution, `stats-reviewer` for statistics, `meteo-reviewer` for meteorology and `scicomm-reviewer` for science communication. They push back, rank concerns and rewrite nothing.
How does handoff work?
`/handoff` creates or updates `.ai/HANDOFF.md` for the next agent or session and `/pickup` reads that state. `/recover-sessions` finds closed Claude Code sessions from on-disk transcripts.
Does it need external service setup?
Not for most of the set. Zotero writes, the Notion work journal, Slack alerts and session names need setup, described in `docs/configuration.md`.
What does /disclose do?
It estimates AI involvement in a deliverable on a 0 to 10 scale, asks you to confirm or revise the estimate and emits a disclosure line naming the model role. Entries go to `.ai/disclosures.jsonl`.
Discussion
No questions or reviews yet.
Related
PluginsAround 200 skills for Claude Code on bioinformatics, chemistry and biostatistics, from RNA-seq analysis to molecular docking
43 skills for Claude Code along the research cycle: surveys, experiments, topic models, reporting and manuscript review
StatsClaw assembles nine Claude Code agents to build statistical packages from a paper or formulas with isolated pipelines.
MCP server that connects Claude Code and Codex to an open R or Python session in RStudio and Positron, where your data lives
Claude Code skills for empirical economics: code and replication audit, bibliography checks, PDF reading and slide decks.
A set of commands, skills and subagents for Claude Code that runs data analysis like a careful team analyst.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| ai-science-toolkit (this card) | no | 83 | |
| Supervisor-Skills | no | 12,379 | twelve skills for research work: idea evaluation, literature review and paper text |
| EasyPlot | no | 114 | a scientific plotting skill in R and Python with no reviewers or handoff |
README in short
The page presents the set as a toolkit for AI-assisted scientific computing, to adopt whole or in parts. The skill and agent tables are generated from their frontmatter. Installation through the plugin or `sync.sh`, the `ai-sessions` function that lists running sessions and a platform support section follow.