AI Research Skills plugin
A marketplace of five Claude Code plugins for researchers: literature search, research design, project memory and manuscript work
- 604 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The zotero-skills plugin creates, edits, deletes and merges entries in your Zotero library
- The delegation plugins run Codex and Antigravity commands on your system, and the Python package searches the network and writes files
- Project state and drafts are saved in `.research/` and `.paper/` files inside your repository
Install
- Works in
Manual install
claude plugin marketplace add WenyuChiou/ai-research-skillsAdds the marketplace. Then `claude plugin install research-workspace@ai-research-skills` installs the base 12 skills.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The zotero-skills plugin creates, edits, deletes and merges entries in your Zotero library
- The delegation plugins run Codex and Antigravity commands on your system, and the Python package searches the network and writes files
- Project state and drafts are saved in `.research/` and `.paper/` files inside your repository
This is third-party code. Review the repository files before installing.
What it does
The repository assembles 17 portable skills into an eight-stage researcher path: from literature search and gap finding to the manuscript and the reply to reviewers. The skills themselves live in five separate repositories; this one holds the Claude Code marketplace, documentation and the schemas that pass results between stages. The main plugin, research-workspace, carries 12 skills: an orchestrator that pauses at human decisions, literature search and comparison, a paper triage matrix, NotebookLM brief verification, research design, project context compression and task routing between agents. The other plugins are added as needed. academic-writing-skills runs the paper outline and draft and audits evidence, zotero-skills creates, edits and deletes Zotero entries, and codex-delegate and antigravity-delegate hand mechanical work to Codex and Antigravity commands. Three ideas keep it steady. Research state lives in `.research/` and `.paper/` files, so a new session reads them instead of needing a recap. Results between stages follow YAML schemas: a claim without a source gets the status gap and cannot count as supported. Automation stops where a human decides: contract changes, skill removal, writes to outside services, accepting meaning and release.
Who it is for. For graduate students, postdocs and research support staff, and for people building agent harnesses who want to see how state, schemas and approval points work in a research process.
Good fit when
- You need to build a literature review, compare the papers found and keep their contradictions
- After dozens of papers you need to learn whether a gap is open and worth closing in the time you have
- Every new agent session needs a recap and you want a persistent project state manifest
- A paper has to be written with claims tied to sources and omissions checked before submission
- You want to manage a Zotero library through an agent or hand heavy mechanical work to Codex
Not a fit when
- You want a finished paper in one action: the set walks you through stages and stops at human decisions
- Your field is far from water resources and agent-based modeling: the author did not validate it on social sciences, machine learning or clinical writing
- You need a guarantee of correctness on real data: behavior is the responsibility of the individual plugin repositories, and the catalog does not cover it
- You do not want to run a Python environment for search automation: without it the plugins work only as instructions
Example request
Build a literature triage matrix for human-in-the-loop research agents, verify identifiers, preserve contradictions and show me the evidence gapsLimitations
The set was assembled and tested by one graduate student and is not validated at corpus scale. It leans toward water resources and agent-based modeling; suitability for social sciences, machine learning and clinical writing was not checked. Correctness on real data rests with the plugin repositories, while this catalog holds only the registry, documents and images.
Plugin install through the marketplace is checked structurally, and the full install-and-trigger loop is verified by the author between releases. Portability was audited on 14 skills: 11 passed with no edits and 3 needed path edits that have since landed. In Hermes the install of one skill was verified and the inference loop was not run. For OpenClaw the structure matches, but there is no full check. Search automation needs the Python package `research-hub-pipeline`, while Zotero, Obsidian and NotebookLM are connected optionally.
How to disable. Remove a plugin with `claude plugin uninstall research-workspace@ai-research-skills` and swap the name for others. The marketplace is removed with `claude plugin marketplace remove ai-research-skills`. The `.research/` and `.paper/` files stay in the project, so delete them yourself.
FAQ
How do I install AI Research Skills in Claude Code?
Add the marketplace with `claude plugin marketplace add WenyuChiou/ai-research-skills`, then run `claude plugin install research-workspace@ai-research-skills`. That is the base set of 12 skills; add the other plugins as needed.
Which plugins are in the marketplace?
Five: research-workspace with 12 skills, academic-writing-skills for the manuscript, zotero-skills for the library, and codex-delegate and antigravity-delegate for handing mechanical work to other CLIs.
Do the skills work outside Claude Code?
Yes, as SKILL.md files. In Codex CLI, Cursor and Windsurf you put them in the rules or skills directory, and in Gemini CLI you paste the text into context. Auto-triggering and the plugin install command are Claude Code only.
What does the research-hub-pipeline package do?
It is the executable runtime for literature search and resumable workflows. Install it with `pip install research-hub-pipeline`, then run `research-hub setup` and `research-hub doctor`. Without it the skills work as instructions.
Will the agent invent sources that do not exist?
The set's schemas account for that: a claim with an empty evidence list must carry the status gap and a reason, so it cannot be passed off as supported. You can ask for identifier checks directly in the literature matrix request.
Where does the set pause for a human decision?
At five points: contract changes, skill removal, writes to outside services, accepting the meaning of a result and release. Reads and reversible steps run automatically.
Discussion
No questions or reviews yet.
Related
PluginsA Claude Code skill for scientific literature reviews: an interview, Scopus and OpenAlex search, DOI checks and a final docx file
Seven Codex skills carry research from topic choice and paper reading to reviewer replies and an editable presentation.
Claude Code skills for the academic pipeline: research, paper drafting, peer review and citation checks
A library of 400+ academic skills: literature search, reviews, statistics, scientific writing and grant proposals
The paper-spine skill takes a topic, materials and data through a scientific paper: literature, text, figures, review and Word, LaTeX and PDF files
A research loop that keeps evidence: a literature review, paper review or survey runs in steps with checks and resume.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| AI Research Skills (this card) | no | 604 | |
| PaperOrchestra | no | 697 | a multi-agent pipeline from outline and plots to a paper, without a plugin registry |
| Humanities Writing Companion | no | 478 | walks a humanities paper through 12 modes without literature search or project management |
| Auto-Empirical Research Skills | no | 4,929 | plugins for empirical social science: causal inference and econometrics |
README in short
The description, in English and Traditional Chinese, opens with a why-what-how table and diagrams of the architecture and states. Install steps by host follow, then five researcher pain points, eight stages with inputs and outputs, sample results and a compatibility table. Limitations close it.