research-paper

Three agent skills for academic research, in one repo. Find real papers, read any paper as a visual experience, and write new papers grounded in real sources. Runtime-neutral — works with 50+ AI coding agents.

npx skills add aniketkrs/research-paper

After install, in any compatible agent session:

/research "graph neural networks for fraud detection" --style ieee
/find-paper "retrieval-augmented generation" --years last-3
/read-paper https://arxiv.org/abs/1706.03762

Date freshness: every paper run starts by checking today's actual date (via date -u +%Y-%m-%d or runtime context). Year-range flags like --years last-3 are computed from today, not from the model's training cutoff. The skill never silently uses stale data — see instructions/freshness.md in any of the three skills.

What this is

Three small, focused skills that work together — or independently — to make academic-paper work fast, rigorous, and not boring.

	Skill	One-line job
✍️	`research-paper`	Writes new papers — research papers, lit reviews, theses, whitepapers, surveys, policy briefs
🔎	`get-research-paper`	Finds real existing papers — searches arXiv, Scholar, PubMed, Semantic Scholar
📖	`read-research-paper`	Reads any paper (URL / arXiv / DOI / PDF / DOCX / PPTX / image) and renders it visually

All three are runtime-neutral. They install with one command and work with Claude Code, OpenCode, Cursor, Cline, Codex, Aider, Amp, Antigravity, AiderDesk, Augment, IBM Bob, and 50+ other agent runtimes via npx skills.

Install

One command — all three skills, no prompts

npx skills add aniketkrs/research-paper --yes --skill '*'

The --yes --skill '*' flags install all three skills in one shot without prompting. The installer auto-detects every agent runtime on your machine and places the skills in the universal .agents/skills/ directory.

Verify

npx skills list

Expected: research-paper, get-research-paper, read-research-paper all listed.

Other install options

# Install only one skill
npx skills add aniketkrs/research-paper --yes --skill research-paper

# Pin to a version
npx skills add aniketkrs/research-paper#v2.4.0 --yes --skill '*'

# User-scope (global) instead of project-scope
npx skills add aniketkrs/research-paper --yes --skill '*' --global

# Direct from GitHub (no skills CLI; this installer also installs all skills by default)
npx -y github:aniketkrs/research-paper install

For per-platform manual install (Claude Desktop, claude.ai web, Anthropic API/SDK, etc.), see INSTALLATION.md.

The three skills

✍️ research-paper — writes new papers

Produces publication-ready research papers, literature reviews, theses, whitepapers, surveys, and policy briefs.

Trigger it:

/research "topic" --style ieee --depth comprehensive
/literature-review "topic" --systematic --sources 50
/whitepaper "topic" --audience technical
/thesis "Chapter 3: Methodology" --style harvard
/policy "topic" --depth standard

…or in plain English: "write a research paper on graph neural networks for fraud detection".

Produces:

10 paper formats: arXiv, IEEE, ACM, Nature, Harvard, lit-review, thesis chapter, whitepaper, survey, policy brief.
7 citation styles: Harvard, APA, IEEE, MLA, Chicago, Nature, arXiv-numeric. Switchable with one flag.
Real visualizations: charts, tables, heatmaps, flowcharts, PRISMA diagrams, forest plots.
Methodology section with sample-size justification and validity threats.
Statistical validation with effect sizes, CIs, multiple-comparison correction.
Three-persona simulated peer review (methodologist + domain expert
- reader).
Plain-English summary alongside the technical content.

What it won't do: invent citations, DOIs, or coauthors. Unverifiable items are flagged [UNVERIFIED] and surfaced in Known-gaps.md.

🔎 get-research-paper — finds papers on a topic

Searches arXiv, Google Scholar, PubMed, Semantic Scholar, DBLP, ACM DL, IEEE Xplore, and OpenReview. Returns a curated reading list with verified DOIs, key findings, and ready-to-cite metadata.

Trigger it:

/get-research-paper "topic" --n 25 --years 2020-2024 --depth deep
/find-paper "topic"
/papers-on "topic"
/scholar "topic"

…or: "find research papers on retrieval-augmented generation".

Produces:

A ranked reading list (reading-list.md) with quality scores.
A canonical bibliography.yaml ready to feed the writer skill.
A 1–3 paragraph field briefing.
Source-quality scoring (authority + rigor + recency).
Diversity heuristics: per-author cap, per-venue cap, ≥1 review, ≥1 foundational paper.
DOI verification + retraction screening (when web tools are available).

📖 read-research-paper — reads any paper, makes it not boring

Take any paper input — arXiv URL, arXiv ID, DOI, PDF, DOCX, PPTX, image, plain text — and render it as a visual reading experience.

Trigger it:

/read-research-paper https://arxiv.org/abs/1706.03762
/read-paper ./paper.pdf
/explain-paper https://doi.org/10.1145/3589334
/visualize-paper ./slides.pptx
/tldr-paper ./scan.png

…or: "read this research paper [URL]", "explain this paper", "make this paper visual".

Produces a multi-layer Markdown rendering with:

One-page infographic at the top (mind map + headline numbers).
TL;DR (5–8 sentences).
Plain-English summary (5–10 sentences).
Section-by-section walk-through with plain-English alongside the technical content (not replacing it).
Method flowchart (Mermaid).
Key-findings infographic (matplotlib when available, Markdown otherwise).
Comparison table to baselines.
Related-work timeline.
"Why this matters" footer.
Verification trail.

Three-tier "don't bluff" cascade: local cache → live fetch → bundled corpus → model knowledge with [UNVERIFIED] flags. Source tier is always declared in the output footer.

Supported file formats

Read (input)

The read-research-paper skill handles any of these as input:

Format	Extensions	Always available?
Markdown, plain text, JSON	`.md`, `.markdown`, `.txt`, `.json`	✅
LaTeX	`.tex`, `.latex`	✅
HTML	`.html`, `.htm`	✅ (with regex fallback)
CSV / TSV	`.csv`, `.tsv`	✅ (basic)
RTF	`.rtf`	✅ (with regex fallback)
PDF	`.pdf`	install `pdfplumber` or `pypdf`
DOCX	`.docx`	install `python-docx`
PPTX	`.pptx`	install `python-pptx`
XLSX	`.xlsx`, `.xls`	install `pandas` + `openpyxl`
EPUB	`.epub`	install `ebooklib`
Images (OCR)	`.png`, `.jpg`, `.tiff`, `.bmp`	install `pytesseract` + Tesseract

One command to enable everything:

pip install pdfplumber python-docx python-pptx pandas openpyxl \
            beautifulsoup4 striprtf ebooklib pytesseract Pillow

Write (output)

The research-paper skill produces Markdown by default. To convert to other formats:

Format	Extension	Renderer
Markdown	`.md`	native (always)
HTML	`.html`	Pandoc
DOCX	`.docx`	Pandoc
LaTeX	`.tex`	Pandoc
PDF	`.pdf`	Pandoc + LaTeX engine
RTF	`.rtf`	Pandoc
EPUB	`.epub`	Pandoc
ODT	`.odt`	Pandoc
PPTX	`.pptx`	Pandoc

Install Pandoc (and optionally LaTeX for PDFs):

# macOS
brew install pandoc                      # base
brew install --cask mactex-no-gui         # for PDF output

# Linux
apt install pandoc                        # base
apt install texlive                        # for PDF output

# Windows
choco install pandoc                       # or scoop install pandoc
choco install miktex                        # for PDF output

Then:

/research "topic" --output paper.pdf
/research "topic" --output paper.docx

Self-test to see what's available on your machine:

python skills/read-research-paper/toolchains/read_any_file.py --self-test
python skills/research-paper/toolchains/convert_output.py --self-test

Full reference: skills/read-research-paper/sources/file-formats.md.

End-to-end workflow

The three skills chain cleanly:

┌─────────────────────────────────────────────────────────────┐
│   /find-paper "topic"                                        │
│   └─→ reading-list.md + bibliography.yaml                    │
└─────────────────────────────┬───────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│   /read-paper <URL>      (visualize ANY single paper)         │
│   └─→ paper-visual.md  (mind map + flowchart + plain English) │
└─────────────────────────────┬───────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│   /research "topic" --bibliography ./topic/bibliography.yaml  │
│   └─→ paper-final.md → convert_output → paper.pdf / .docx    │
└─────────────────────────────────────────────────────────────┘

Or use any one standalone — they don't require each other.

Quick start

Find papers on a topic

/find-paper "transformer architectures for time-series forecasting" --n 10

The skill returns a curated reading list with verified DOIs, key findings per paper, and a bibliography.yaml ready for the writer skill.

Read a single paper visually

/read-paper https://arxiv.org/abs/1706.03762

The skill returns a multi-layer Markdown rendering with mind map, flowchart, infographic, and plain-English alongside the technical content. Cached locally so re-asking is instant.

Write a paper using a curated bibliography

/research "predicting customer churn with graph neural networks" \
    --style ieee \
    --depth comprehensive \
    --bibliography ./gnn-fraud-detection/bibliography.yaml

The skill walks: plan → literature review → methodology → analysis → visualization → drafting → citations → validation → review → ship.

Convert the result to PDF / DOCX

python skills/research-paper/toolchains/convert_output.py \
    --input ./gnn-fraud-detection/paper-final.md \
    --to pdf \
    --out paper.pdf

What "done" means

A paper is delivered only after passing every quality gate:

≥ 1500 words (configurable).
≥ 8 references with verified DOIs.
≥ 1 figure or table.
All required sections present (abstract, intro, methodology, results, discussion, limitations, conclusion, references).
Plain-English summary present.
Reproducibility statement present.
Future Work section present.
Academic-quality rubric mean ≥ 4 / 5.
All three reviewer personas score ≥ 3.0.
All [CITATION NEEDED] and [UNVERIFIED] flags resolved or surfaced in Known-gaps.md.

Failures surface in Known-gaps.md. Never silent.

Compatibility

npx skills add auto-detects and installs into 50+ agent runtimes:

Universal	Symlink-supported
Amp, Antigravity, Cline, Codex, Cursor, +10 more	AiderDesk, Augment, IBM Bob, Claude Code, OpenCode, +35 more

Run npx skills add aniketkrs/research-paper --list to see the full list of agents detected on your machine.

Repo structure

research-paper/
├── README.md, LICENSE, CHANGELOG.md, INSTALLATION.md, package.json
├── bin/install.js               ← direct npx installer
├── tests/test-runner.js         ← 116/116 tests pass
├── docs/                        ← architecture, design decisions, FAQ
└── skills/
    ├── research-paper/          ← writes papers (106 files)
    ├── get-research-paper/      ← finds papers (20 files)
    └── read-research-paper/     ← reads any paper (26 files)

Each skill is self-contained — manifest, instructions, prompts, templates, schemas, toolchains, examples, sources, tests.

Optional Python toolchain

The skills work with just filesystem read/write. They emit Markdown tables and Mermaid diagrams as the always-available default.

To enable real charts, statistical validation, multi-format file I/O, and Pandoc output:

# Charts + statistical validation (research-paper)
pip install pandas numpy scipy statsmodels matplotlib seaborn pyyaml

# Multi-format file reading (read-research-paper)
pip install pdfplumber python-docx python-pptx openpyxl \
            beautifulsoup4 striprtf ebooklib pytesseract Pillow

# Multi-format output (research-paper) - install Pandoc + LaTeX system-wide
# macOS:    brew install pandoc && brew install --cask mactex-no-gui
# Linux:    apt install pandoc texlive
# Windows:  choco install pandoc miktex

When any tool is missing, the skill detects it and falls back — never silent failure, always a clear "install X to enable" message.

Extending

The skills are modular. Each one is a folder with a clear extension boundary.

Add a new venue (e.g., LNCS): drop a template in templates/<venue>.md + register in manifest.json.
Add a new citation style: extend citation_engine/citation-styles.md and toolchains/format_bibliography.py.
Add a new chart type: extend visualization_engine/decision-engine.md and toolchains/generate_charts.py.
Add a new file format: extend read-research-paper/toolchains/read_any_file.py and read-research-paper/sources/file-formats.md.

Full extension guide: docs/extending.md.

Tests

Lightweight test suite in tests/test-runner.js:

node tests/test-runner.js

Covers:

File structure (every required file present).
SKILL.md frontmatter (valid YAML).
manifest.json (valid JSON, required fields).
JSON Schemas (paper-schema, citation-schema, visual-paper schema, etc.).
Python toolchain self-tests (graceful degradation).
Citation pipeline smoke test against fixtures.

116/116 tests pass.

Troubleshooting

Symptom	Likely cause	Fix
Skill never activates	Restart your agent session after install
Charts come out as Markdown only	Python plotting deps missing	`pip install pandas matplotlib seaborn`
Can't read a `.pdf`	PDF library missing	`pip install pdfplumber`
Can't read a `.docx`	DOCX library missing	`pip install python-docx`
Can't OCR an image	OCR not installed	`pip install pytesseract Pillow` + install Tesseract
Can't convert to PDF	Pandoc / LaTeX missing	install Pandoc + a LaTeX engine
Skill activates for unrelated requests	Trigger patterns too broad	edit `manifest.json → trigger.patterns`
Output truncated mid-section	Context pressure	switch to multi-file output (`--depth comprehensive`)

License

MIT — see LICENSE.

Contributing

Issues and PRs welcome at github.com/aniketkrs/research-paper.

When filing an issue, please include:

Your runtime (Claude Code, OpenCode, etc.)
The slash command or prompt that triggered the issue
The contents of Known-gaps.md if any
The relevant validation-report.md if any

Versioning

Version	Highlights
2.3.0	Multi-format file I/O: read PDF/DOCX/PPTX/XLSX/EPUB/images; write PDF/DOCX/HTML/LaTeX/EPUB/RTF/ODT/PPTX
2.2.0	Added `read-research-paper` skill (visual paper reading)
2.1.0	Added `get-research-paper` skill (paper discovery)
2.0.x	Initial release of `research-paper` skill (paper writing)

See CHANGELOG.md for the full history.

Navigation Menu

Repository navigation

About

Repository files navigation

research-paper

What this is

Install

One command — all three skills, no prompts

Verify

Other install options

The three skills

✍️ research-paper — writes new papers

🔎 get-research-paper — finds papers on a topic

📖 read-research-paper — reads any paper, makes it not boring

Supported file formats

Read (input)

Write (output)

End-to-end workflow

Quick start

Find papers on a topic

Read a single paper visually

Write a paper using a curated bibliography

Convert the result to PDF / DOCX

What "done" means

Compatibility

Repo structure

Optional Python toolchain

Extending

Tests

Troubleshooting

License

Contributing

Versioning

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
bin		bin
docs		docs
skills		skills
tests		tests
.gitignore		.gitignore
CHANGELOG.md		CHANGELOG.md
INSTALLATION.md		INSTALLATION.md
LICENSE		LICENSE
README.md		README.md
package.json		package.json

Latest commit

History

Folders and files

Repository files navigation

research-paper

What this is

Install

One command — all three skills, no prompts

Verify

Other install options

The three skills

✍️ research-paper — writes new papers

🔎 get-research-paper — finds papers on a topic

📖 read-research-paper — reads any paper, makes it not boring

Supported file formats

Read (input)

Write (output)

End-to-end workflow

Quick start

Find papers on a topic

Read a single paper visually

Write a paper using a curated bibliography

Convert the result to PDF / DOCX

What "done" means

Compatibility

Repo structure

Optional Python toolchain

Extending

Tests

Troubleshooting

License

Contributing

Versioning

About

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages