An open API service indexing awesome lists of open source software.

https://github.com/scdenney/open-science-skills

A library of Agentic Skills for Claude Code and LLM agents, codified from social science standards and methodological texts for doing experimental and other kinds of work.
https://github.com/scdenney/open-science-skills

anthropic claude-code claude-code-skills conjoint experimental-research fair list-experiment literature-review manuscript-qa methods-reporting open-science peer-review pre-registration social-science survey-experiments text-classification topic-modeling

Last synced: 1 day ago
JSON representation

A library of Agentic Skills for Claude Code and LLM agents, codified from social science standards and methodological texts for doing experimental and other kinds of work.

Awesome Lists containing this project

README

          


Open Science Skills — vintage typewriter, globe, and books labeled Open Access, Collaboration, Transparency, Reproducibility, beneath a framed title sign.

# Open Science Skills

[![Claude Code](https://img.shields.io/badge/Claude_Code-plugin-D97757?logo=anthropic&logoColor=white)](https://code.claude.com/docs/en/skills)
[![OpenAI Codex](https://img.shields.io/badge/OpenAI_Codex-library-111111?logo=openai&logoColor=white)](codex/README.md)
[![version](https://img.shields.io/badge/version-2.12.0-blue)](https://github.com/scdenney/open-science-skills/releases)
[![license](https://img.shields.io/badge/license-CC%20BY--NC%204.0-lightgrey)](LICENSE)
[![skills](https://img.shields.io/badge/skills-35-blue)](#skills)
[![updated](https://img.shields.io/badge/updated-July%202026-green)](https://github.com/scdenney/open-science-skills/commits/main)
[![sources](https://img.shields.io/badge/sources-150%2B-purple)](SOURCES.md)
[![PRs welcome](https://img.shields.io/badge/PRs-welcome-brightgreen)](#contributing)

Reusable agent workflows for experimental social science, computational text analysis, manuscript QA, and transparent reporting. The same research methods are packaged natively for both Claude Code and OpenAI Codex.

| Platform | Package | Invoke |
|---|---|---|
| [Claude Code](https://code.claude.com/docs/en/skills) | 35 skills in the [`oss` plugin](plugin/skills) | `/oss:skill-name` |
| [OpenAI Codex](https://developers.openai.com/codex/skills) | 34 skills in the [`codex/` library](codex/README.md) | `$skill-name` |

The two libraries track each other closely. On the Codex side there is no `presubmit`.

This is the toolkit I use in my own research. It is built from a curated corpus of methodology texts and grows as I add new sources, ideas, and skills. Authoring and editing are mine, with help from Opus 4.8, Gemini 3.0, and ChatGPT 5.4.

Both libraries follow the official authoring guidance for [Claude Code](https://platform.claude.com/docs/en/agents-and-tools/agent-skills/best-practices) and [Codex](https://developers.openai.com/codex/skills): procedural workflows over definitions, with trigger-rich descriptions and progressive disclosure. The methods stay aligned across platforms; only invocation and tooling adapt to each runtime.

> These skills support the research and writing process. They do not replace it. They follow APSA, JARS, DA-RT, TOP, and FAIR open-science expectations, and all guidance is grounded in 150+ published sources and documented workflow patterns. See [**SOURCES.md**](SOURCES.md) for the full bibliography.

---

## Contents

[Skill Map](#skill-map) · [How Skills Work](#how-skills-work) · [Installation](#installation) · [Skills](#skills) · [Contributing](#contributing) · [License](#license)

**Skills:** [Project Setup](#project-setup) · [Ideation](#ideation) · [Research Design](#research-design) · [Analysis](#analysis) · [Corpus Processing](#corpus-processing) · [Writing & Reporting](#writing--reporting) · [Figures & Tables](#figures--tables) · [Manuscript QA](#manuscript-qa) · [Review & Submission](#review--submission)

---

## Skill Map

```mermaid
flowchart LR
A[Research Design] --> B[Analysis]
A --> C[Writing and PAP]
B --> D[Methods Reporting]
C --> D
D --> E[Manuscript QA and FAIR]
E --> F[Review and Submission]

G[Ideation] --> A
G --> C

A -.-> A1[conjoint / survey / list / cross-national]
B -.-> B1[topic modeling / text classification / council voting / model committee / calibration / OCR + OCR eval]
C -.-> C1[hypotheses / literature review / narrative / preregistration]
D -.-> D1[figures / tables / methods reporting]
E -.-> E1[FAIR / citations / fact-check / figure-table-audit / replication-package / archive checks]
F -.-> F1[paper-tex / paper-review-lite / paper-review-lite-codex / presubmit / journal-review]
G -.-> G1[diverge / diverge-codex / fable-orchestrate or 46-orchestrate]
```

Use the domain skills when designing or analyzing a study. Use the manuscript-QA skills when a draft exists and you need to check whether FAIR availability, citations, figures, tables, reporting, and replication materials can survive review.

---

## How Skills Work

Invocation depends on the platform:

| Platform | Implicit | Explicit | Source |
|---|---|---|---|
| Claude Code | Loads matching skills from prompt context | `/oss:skill-name` | [`plugin/skills/`](plugin/skills) |
| Codex | Loads matching skills from their descriptions | `$skill-name` | [`codex/`](codex) |

Orchestration and delegated-review variants require explicit invocation because they start subagents or external model calls.

---

## Installation

### Claude Code

#### Option 1 — Plugin (recommended, installs all skills + slash commands)

**Permanent install** (user-wide, persists across all projects):

```bash
# Step 1: Register the marketplace (one-time)
claude plugin marketplace add scdenney/open-science-skills

# Step 2: Install the plugin
claude plugin install oss@open-science-skills

# Project-only install
claude plugin install oss@open-science-skills --scope project
```

The plugin's slash-command prefix is `oss:` (short for **o**pen **s**cience **s**kills). The marketplace and GitHub repo are still named `open-science-skills`.

**Session-only** (no install required, active for the current session):

```bash
git clone https://github.com/scdenney/open-science-skills.git
cd open-science-skills && claude --plugin-dir ./plugin
```

All 35 skills auto-trigger based on your prompts. All 35 slash commands (`/oss:research-repo`, `/oss:conjoint-design`, `/oss:fair-check`, `/oss:figures`, `/oss:tables`, `/oss:paper-tex`, `/oss:figure-table-audit`, `/oss:replication-package`, and so on) are immediately available. The prefix can be omitted when no other installed plugin claims the same name.

Option 2 — Selective install (choose specific skills, auto-trigger only)

Use the interactive install script to pick only the skills you want:

```bash
git clone https://github.com/scdenney/open-science-skills.git
cd open-science-skills
bash plugin/scripts/install.sh
```

The script lists available skills and lets you choose interactively. Skills are installed to `./.claude/skills/` by default (current project only). Options:

```bash
# Install to user-wide skills directory (all projects)
bash plugin/scripts/install.sh --target ~/.claude/skills

# Install specific skills non-interactively
bash plugin/scripts/install.sh --skill conjoint-design survey-design list-experiment

# Install all skills
bash plugin/scripts/install.sh --all --target ~/.claude/skills
```

Restart Claude Code after installing to load the new skills.

Option 3 — Manual copy (single skill, auto-trigger only)

```bash
git clone https://github.com/scdenney/open-science-skills.git

# Project-level (current project only) — copy the whole skill folder:
# many skills ship reference/, assets/, or scripts/ files their SKILL.md points at
mkdir -p your-project/.claude/skills
cp -R open-science-skills/plugin/skills/conjoint-design \
your-project/.claude/skills/

# User-wide (all projects)
mkdir -p ~/.claude/skills
cp -R open-science-skills/plugin/skills/list-experiment ~/.claude/skills/
```

> **Note:** Manual install gives you auto-trigger only. Slash commands (`/skill-name`) require the plugin.

### Codex

Codex discovers repository skills under `.agents/skills` and user-wide skills under `~/.agents/skills`. From the repository root, install all 34 native skills user-wide:

```bash
mkdir -p "$HOME/.agents/skills"
for skill in "$PWD"/codex/*/; do
ln -sfn "${skill%/}" "$HOME/.agents/skills/$(basename "$skill")"
done
```

For selective and repository-scoped installation, plus the compact Codex catalog, see [`codex/README.md`](codex/README.md).

---

## Skills

The detailed catalog shows Claude Code commands by default. Platform-specific entries are labeled: **Codex** means a Codex-native skill, while **Claude Code → Codex** means a Claude Code skill that calls Codex. Unmarked research workflows also have counterparts in the [Codex catalog](codex/README.md), except for `presubmit`.

### Project Setup

Skill
Invoke
What it does

research-repo
/research-repo
Scaffold or audit a research project repository organized around its source library. For a new repo, build the sources/{og,md,unprocessed} + references.bib spine (PDF to Markdown via OpenDataLoader PDF), a process-source intake command, CLAUDE.md/AGENTS.md, .gitignore, and the archetype-appropriate analysis/manuscript/review folders, then smoke-test the pipeline; for an existing repo, audit against the convention with detection recipes for orphan PDFs, bib drift, and naming. Pairs with process-source (per-PDF intake) and replication-package (publication output).

### Workflow & Orchestration

Skill
Invoke
What it does

fable-orchestrate
Claude Code
/fable-orchestrate
Multi-model orchestration with Fable 5 as tech lead. Routes reasoning-heavy work to a deep-reasoner subagent (Opus), mechanical work to a fast-worker subagent (Sonnet), and fresh-perspective or high-stakes work to Codex, a different-vendor GPT-5 peer. High-stakes tasks run Opus and Codex in parallel, then synthesize. Ships the two agent definitions and a codex-peer.sh driver.

46-orchestrate
Codex
$46-orchestrate
Sol-led Codex orchestration for complex work. Routes bounded research, implementation, and verification to role-based subagents; uses blind parallel review for high-blast-radius work that is difficult to verify; and keeps planning, integration, and final accountability with the lead.

### Ideation

Skill
Invoke
What it does

diverge
/diverge
Brainstorm-then-select: before implementing, generate 3–5 conceptually distinct approaches labeled by creativity dimension ([Novel], [Surprising], [Diverse], [Conventional]) and hold for selection. Resists defaulting to the most obvious solution. After Creative Preference Optimization (Ismayilzada et al., 2025).

diverge-codex
Claude Code → Codex
/diverge-codex
Cross-model variant of diverge. Delegates the brainstorm to Codex (GPT-5.4) via codex exec, presents its approaches for selection, then has Codex implement the chosen one. A second model family widens the space of approaches.

### Research Design

Skill
Invoke
What it does

conjoint-design
/conjoint-design
Attribute architecture, AMCE/AMIE estimation, power analysis (cjpowR), BART heterogeneity detection, design variants

conjoint-diagnostics
/conjoint-diagnostics
Diagnostic checklist: design, estimation, measurement error (IRR), external validity, interpretation

conjoint-cleaning
/conjoint-cleaning
Qualtrics export to analysis-ready format: column conventions, reshaping, choice mapping, translation, pilot detection, validation

survey-design
/survey-design
Question construction, scale design, survey flow, pretesting, respondent burden, social desirability mitigation

cross-national-design
/cross-national-design
Cross-national survey experiments: per-country power, sensitivity bias auditing, instrument localization

list-experiment
/list-experiment
Item count technique: pre-design sensitivity assessment, control list design, NLSreg/MLreg estimation, assumption testing, placebo diagnostics

### Analysis

Skill
Invoke
What it does

topic-modeling
/topic-modeling
STM with metadata covariates, topic count selection via coherence-exclusivity diagnostics, reporting

text-classification
/text-classification
LLM-based classification: codebook design, learning regime selection, human-LLM hybrid workflows, validation

model-council-voting
/model-council-voting
Panel of models as independent coders: consensus rules, chance-corrected agreement (Cohen/Fleiss/Krippendorff), correlated-error diagnostics, validation beyond the panel

model-committee
Claude Code + Codex
/model-committee
$model-committee
GPT-5.5 and Claude Opus 4.8 as a deliberative committee: use-case gate, blind proposals, cross-critique and revision, blinded cross-ranking, and a precommitted rule for one decision

llm-calibration-logprobs
/llm-calibration-logprobs
Within-model confidence from token log-probabilities: aggregation, triage, calibration (ECE, Brier, reliability diagrams) against human ground truth

### Corpus Processing

Skill
Invoke
What it does

vlm-ocr-pipeline
/vlm-ocr-pipeline
VLM-based OCR: model selection, image handling, prompt engineering, batch strategy, accuracy evaluation, reproducibility

post-ocr-cleanup
/post-ocr-cleanup
Post-OCR cleanup: LLM and rule-based correction, quality diagnostics, multilingual considerations, corpus-level QA, provenance

vlm-ocr-evaluation
/vlm-ocr-evaluation
Compare OCR systems before bulk runs: candidate set, stratified ground truth, CER/WER with declared normalization, per-language and per-stratum accuracy

### Writing & Reporting

Skill
Invoke
What it does

hypothesis-building
/hypothesis-building
Falsifiability, counterfactuals, DAGs, FPCI, three-level hypothesis specification, equivalence testing, SESOI

literature-review
/literature-review
Evidence maps, closest-prior-work assessment, gap verdicts, literature clusters, synthesis plans

narrative-building
/narrative-building
Introduction logic, literature reviews, the "Why-to-If-Then" funnel, cumulative framing, multi-experiment coherence

pre-registration-writing
/pre-registration-writing
PAP structure, registry selection, analytical strategy specification, code pre-registration, deviation documentation

methods-reporting
/methods-reporting
40-item reporting checklist: CONSORT standards, JARS preregistration elements, DA-RT transparency

paper-tex
/paper-tex
Typeset a working paper or journal submission in house-style LaTeX from any draft (Markdown, Word, TeX, ODT, HTML): pandoc conversion, EB Garamond template, \figcap title+note captions, [H] floats, single-spaced title block with the introduction on its own page, and journal-specific prep (spacing, page limit, anonymization, disclosures). Bundles a tested convert-and-build driver.

### Figures & Tables

Skill
Invoke
What it does

figures
/figures
Design publication-quality figures: chart choice from comparison, scales, color, legend ordering matched to visual order, self-contained captions, reproducibility

tables
/tables
Design publication-quality tables: column order matching the argument, row grouping, precision and uncertainty conventions, self-contained titles and notes, code-generated workflows

### Manuscript QA

Skill
Invoke
What it does

fair-check
/fair-check
FAIR audit for completed manuscripts: data/code/material availability, repository metadata, persistent identifiers, licenses, access restrictions, reuse conditions

citation-check
/citation-check
In-text/reference parity, DOI and source-status checks, fabrication/existence verification (Crossref/OpenAlex), stale working papers, citation-style and support audits

fact-check
/fact-check
Verify each in-text claim is actually supported by its cited source. Runs citation-check first, then locates each source's knowledge-base Markdown (sources/md/) and audits claim support, overclaiming, direction, scope, and misattribution. Pairs with process-source.

figure-table-audit
/figure-table-audit
End-stage QA of the finished figure/table set: inventory, cross-references, text-to-table consistency, accessibility, SI and replication linkage. Pairs with figures and tables (production-stage).

replication-package
/replication-package
Scaffold or audit a replication package at a target directory. Generates folder structure, README, master.R, figure/table crosswalk, codebook template, LICENSE placeholder, .gitignore, and a pre-release checklist. Platform-neutral (Harvard Dataverse, OSF, Zenodo, GitHub releases). Adapted from Yusaku Horiuchi's replication-package-guide with FAIR-principle integration. Pair with fair-check.

### Review & Submission

Skill
Invoke
What it does

paper-review-lite
/paper-review-lite
Critical-Reviewer-style pre-submission audit. Parallel quote-grounded sub-agents, verification cross-check, CONSORT and pre-reg audit for experimental papers. Single-model, in-session counterpart to the standalone presubmit CLI.

paper-review-lite-codex
Claude Code → Codex
/paper-review-lite-codex
Cross-model adversarial variant of paper-review-lite. Claude and Codex (GPT-5.4) independently apply the same review specification, then each cross-checks the other's findings; surviving issues are annotated by confidence (mutual, asymmetric-confirmed, single-model-adjudicated). Roughly 22 model calls. Use before submission for maximum pressure and a second model family's blind spots.

presubmit
/presubmit
Activator and setup wizard for the standalone presubmit Python CLI. Walks first-time users through install (venv + pip install -e .), Anthropic API key setup, and output location, then runs the heavier 30+ stage adversarial pipeline (resumable, cost-tracked, optional --math and --code-dir add-ons). Heavier API-driven counterpart to paper-review-lite.

journal-review
/journal-review
Drafts a senior-peer referee report on someone else's manuscript for a social-science journal. Five parallel finder sub-agents (Breaker, Butcher, Shredder, Void, Situator), a Blue Team filter, Chief Reviewer synthesis, and Tone Guard sanitization produce a 1,200–2,000-word report (Recommendation, Summary, Major Concerns, Additional Concerns, Suggestions for Revision). Different role from paper-review-lite and presubmit, which are calibrated for self-audit. This one acts as a third-party referee and is meant to support, not replace, your own assessment of a manuscript.

---

## Contributing

PRs welcome. To add a new skill:

1. Create `plugin/skills//SKILL.md` following the [skill authoring best practices](https://platform.claude.com/docs/en/agents-and-tools/agent-skills/best-practices)
2. Add `plugin/commands/.md` (see existing examples — one paragraph activation prompt + `$ARGUMENTS`)
3. Add the Codex-native package at `codex//` with `SKILL.md` and `agents/openai.yaml`, unless the workflow is intentionally platform-specific
4. Add sources to `SOURCES.md` under a new or existing section
5. Update the platform catalogs and badges
6. Run `bash plugin/scripts/check.sh` and validate each changed Codex skill before opening a PR

## License

This project is licensed under [Creative Commons Attribution-NonCommercial 4.0 International](LICENSE). The skills are intended for noncommercial scholarly and educational use.

The `citation-check`, `literature-review`, `figures`, `tables`, and `figure-table-audit` skills remix workflow ideas from [Cheng-I Wu's Academic Research Skills for Claude Code](https://github.com/Imbad0202/academic-research-skills), also licensed CC BY-NC 4.0. The instructions here are rewritten for this repository's open-science and experimental-social-science scope.

The `replication-package` skill adapts the structural conventions in [Yusaku Horiuchi's replication-package-guide](https://github.com/yhoriuchi/replication-package-guide) (the source for single-entry-point, compact vs. build/analyze layouts, figure/table crosswalk, paper-consistency check, correction workflow, and pre-release checklist). FAIR-principle integration and Claude Code/Codex skill packaging are added on top; Harvard Dataverse and other platform-specific upload mechanics are not included. Cite Horiuchi's guide if you publish a package built with this skill.