Examples — Research Agora

Choose Your Domain

Each section contains self-contained prompts you can copy directly into Claude.ai (browser) or Claude Code (CLI). Replace [placeholders] with your content.

If you work with...	Jump to
Data files, CSV/Excel, visualizations, EDA pipelines	Code & Data
Writing abstracts, introductions, discussions, titles	Paper Writing
Papers, citations, literature reviews, BibTeX files	Literature Search & Citation Verification
Email, teaching materials, admin documents, rebuttals	Writing & Admin
R, clinical workflows, C++/TypeScript, econometrics	Technical Workflows
Automated pipelines, Zotero+MCP, privacy audits	Advanced Research

General principles: Verify everything that matters. Start with one task. Rephrase, don't repeat.

Code & Data Visualization

Prompts for exploratory data analysis, publication-quality figures, and data cleaning.

EDA Pipeline

Prerequisites: Claude Code + data file on disk, or Claude.ai + paste/upload data

Try this now:

Load [path/to/data.csv]. Produce an EDA report:

1. Shape, dtypes, missing values per column

2. Distribution plots for all numeric columns

3. Correlation heatmap

4. Top 3 anomalies or unexpected patterns

Save all figures to outputs/ as PDF (vector).

What to verify: Row/column counts match your expectations. Missing value percentages are plausible. Correlation values align with domain knowledge.

Related skills: /code-simplify, /tikz-figures

Publication-Quality Figure

Prerequisites: Claude Code + matplotlib/seaborn installed

Try this now:

Create a publication-quality figure from [path/to/data.csv]:

- Use the Okabe-Ito colorblind-safe palette

- Font size: 10pt for labels, 8pt for ticks

- Export as PDF (vector) at 3.5 inches wide (single-column)

- No title (caption goes in LaTeX)

- Include error bars where appropriate

What to verify: Colors are distinguishable in grayscale. Axis labels are readable at print size. Error bars represent the correct statistic (SD vs. SEM vs. CI).

Related skills: /tikz-figures, /latex-consistency

Paper Writing

Prompts for drafting core paper sections — abstract, introduction, discussion, and titles. Works best when you supply your method, results, and context directly.

Abstract

Prerequisites: Claude.ai or Claude Code — no installation needed

Try this now:

/paper-abstract

My paper: [one-paragraph summary of your method and results]

Venue: [NeurIPS / ICML / ICLR / AAAI / other]

Key result: [single most important finding, with numbers]

Contribution type: [new method / new dataset / analysis / theory]

What to verify: All numbers match your results section exactly. No claims that aren't backed by your experiments. The problem statement matches your introduction.

Related skills: /paper-abstract

Introduction

Try this now:

/paper-introduction

Problem: [what problem does your paper solve, and why does it matter?]

Gap: [what does existing work fail to do?]

Approach: [one sentence describing your method]

Contributions: [3-4 bullet points listing your specific contributions]

Related work to cite: [list key papers you want positioned against]

What to verify: The gap claim is accurate — don't overstate what prior work misses. Contribution bullets are falsifiable. The narrative flows from problem to gap to your solution.

Related skills: /paper-introduction

Discussion Section

Try this now:

/paper-discussion

Main findings: [paste your key results and numbers]

Surprising result (if any): [anything that didn't go as expected]

Limitations: [what your method can't do or where it breaks down]

Broader impact: [what this enables for future work]

What to verify: Limitations are honest and complete — reviewers will add any you omit. Broader impact claims aren't overclaimed. All "future work" suggestions are technically plausible.

Related skills: /paper-discussion

Paper Titles

Try this now:

/paper-abstract

Abstract: [paste your abstract]

Venue: [target venue]

Style preference: [descriptive / catchy acronym / question-form / neutral]

What to verify: The title accurately reflects your actual contribution. If using an acronym, make sure it's not already taken by a prominent paper.

Related skills: /paper-abstract

Literature Search & Citation Verification

Prompts for literature search, synthesis, gap identification, and citation verification — with explicit guards against hallucination.

Warning: Studies report LLMs hallucinate 18–29% of citations in literature reviews. These prompts build in verification from the start.

Methodology-Focused Search

Try this now:

Search for papers that use [specific method] to study [specific problem].

For each paper found:

- Full citation (authors, title, venue, year)

- One-sentence summary of the methodological contribution

- How it differs from [your approach]

If you are uncertain whether a paper exists, write [UNCERTAIN] rather than guessing.

What to verify: Every citation exists. Check at least 3 on Semantic Scholar.

Related skills: /literature-synthesizer, /paper-references

Citation Verification (CLI)

Try this now:

/paper-references

Checks every entry in your .bib file against Semantic Scholar and CrossRef. Flags mismatches, preprints with published versions, and entries that don't resolve to any known paper.

Citation tools comparison:

Tool	What it checks	Notes
SwanRef	Citation existence	Free, batch upload
Semantic Scholar API	Programmatic access	Free API key
Elicit	Structured data extraction	Free tier available

Writing & Admin

Everything here works in Claude.ai with zero installation.

Administrative Email Draft

Try this now:

Draft a professional email:

- From: [your name, your role]

- To: [recipient, their role]

- Purpose: [what you need]

- Tone: [formal/collegial/brief]

- Constraints: [max length, deadline to mention, attachments to reference]

Write the subject line and body. Do not invent facts.

What to verify: No fabricated details. Tone matches your relationship with the recipient. All facts are correct.

Rebuttal Paragraph

Try this now:

Reviewer wrote: "[paste reviewer comment]"

Draft a rebuttal paragraph that:

1. Acknowledges the concern (one sentence)

2. Provides evidence addressing it

3. Describes what changed in the revision

Tone: professional, concessive opening, then firm evidence.

What to verify: The evidence you cite is real. The changes described actually happened. The tone is not defensive.

Related skills: /review-triage

Technical Workflows

Prompts for R/econometrics, clinical/medical imaging setup, C++/TypeScript projects, and referee responses. Assumes terminal familiarity.

R Fixed-Effects Panel Regression

Try this now:

Read [path/to/panel_data.csv].

Estimate a two-way fixed-effects model:

  Y = [outcome] ~ [treatment] + [controls] | [unit_FE] + [time_FE]

Use the fixest package. Cluster standard errors at the [unit] level.

Output a modelsummary table comparing 3 specifications.

Export as LaTeX (booktabs).

What to verify: Number of observations matches your data. Fixed effects are applied to the correct dimensions. Standard errors are clustered at the right level.

Medical/Clinical Setup (Architecture First, No Data)

Try this now:

I'm building a [imaging modality] pipeline for [clinical task].

Do NOT process any patient data.

Design the project architecture:

1. Directory structure (data/, models/, configs/, outputs/)

2. .claudeignore excluding all DICOM, NIfTI, and patient-identifiable files

3. Config template for model parameters

4. A CLAUDE.md that encodes "never read files in data/"

What to verify: The .claudeignore actually blocks all sensitive file types. The CLAUDE.md verification requirements are explicit. No patient data paths are referenced.

Advanced Research

Prompts for connecting external tools (Zotero MCP), auditing your setup, and maintaining AI-generated code.

Zotero + MCP Library Search

Prerequisites: Claude Code + Zotero desktop + Zotero MCP server configured

Try this now:

Search my Zotero library for papers related to [your topic].

For each match, show: title, authors, year, and which collection it's in.

Then identify 3 gaps — topics I should have papers on but don't.

What to verify: The papers returned actually exist in your Zotero library. The "gaps" are genuine, not hallucinated subfields.

Privacy Audit

Try this now:

Audit this session's privacy posture:

List every file you've read in this session

Flag any that contain secrets, credentials, or PII

Check if .claudeignore exists and what it blocks

Report whether DISABLE_TELEMETRY is set

Recommend fixes for any issues found

What to verify: Cross-check the file list against your expectations. Verify the .claudeignore recommendations are complete.

Related skills: /audit-my-setup

Cognitive Debt Audit

Try this now:

Audit [path/to/project] for cognitive debt:

1. Which files were likely AI-generated? (heuristics: uniform style, no TODOs, generic variable names)

2. For each: can a human maintainer understand it without the AI context?

3. Are there tests? If not, which functions most need them?

4. Is there a CONTRIBUTING.md that explains the architecture?

Produce a prioritized refactoring plan. Do NOT refactor — just plan.

What to verify: The AI-generated file detection is plausible but imperfect. Read the refactoring plan before acting on it — it's a starting point, not a prescription.