Skip to content

About

An evidence-first research toolkit for AI agents and human researchers: 64 focused skills for literature discovery, critical reading, systematic reviews, experiment design, ML evaluation, statistics, reproducibility, scientific writing, and peer review—plus local Python utilities, reusable prompts, and templates for rigorous, transparent research.

Topics

Resources

Contributing

Security policy

Stars

3 stars

Watchers

0 watching

Forks

Repository files navigation

Reasoningai.studio

Turn research questions into defensible findings—with evidence you can trace at every step.

An evidence-first research toolkit for AI agents and human researchers: 64 focused skills, 12 reusable prompts, practical research templates, and local Python utilities for more transparent, reproducible research.

ReasoningAI Studio

Research workflows, not generic prompts

ReasoningAI Studio brings research tasks into one coherent, inspectable workflow. Its 64 standalone English-language skills cover the research lifecycle—from framing a question and finding literature to evaluating evidence, designing experiments, auditing machine-learning results, writing a paper, and responding to peer review.

The collection combines 34 ReasoningAI workflows with 30 clearly namespaced ResearchForge Skills, authored by the same project maintainer. All 64 skills are licensed under CC BY 4.0; see NOTICE.md for provenance and the two prompt-file MIT exceptions.

Research workflow

What's included

  • 64 portable agent skills with defined inputs, task-specific steps, concrete outputs, and research-integrity boundaries.
  • 12 copy-ready English prompts for literature synthesis, paper reading, methods review, claim auditing, reproducibility, statistics, and research communication.
  • Local Python utilities for scholarly metadata search, DOI lookup, BibTeX checks, screening ledgers, PRISMA-style count summaries, claim-evidence audits, SHA-256 manifests, and common secret-pattern scans.
  • Research scripts and templates for bibliography verification, dataset auditing, reproducibility preflight, research protocols, evidence matrices, experiment plans, and claim-evidence tracking.
  • Synthetic examples and automated tests so you can inspect the expected inputs and validate the local tools without using private research data.

Browse the complete skill catalog, architecture, and research ecosystem references.

Quick start

Requires Python 3.10 or newer. The local utilities use the Python standard library; no runtime third-party packages are required.

python -m pip install .
python -m reasoningai doctor
python -m reasoningai search "retrieval augmented generation" --provider openalex --limit 10

To use a skill, open skills/<skill-name>/SKILL.md in a compatible AI assistant. Start with the skill catalog and load only the workflow relevant to the task.

Local research utilities

# Audit a clean illustrative BibTeX file
python -m reasoningai bib-audit examples/references.bib

# Demonstrate duplicate-DOI detection (this fixture is expected to report an issue)
python -m reasoningai bib-audit examples/references-with-duplicates.bib

# Verify DOI metadata against Crossref; --offline prevents network requests
python scripts/verify_bibliography.py examples/researchforge-illustrative-references.bib --offline

# Audit a CSV or JSONL dataset for duplicate rows and exact-text split leakage
python scripts/audit_dataset.py examples/researchforge-synthetic-dataset.csv --text-column text --label-column label --split-column split

# Run a heuristic reproducibility preflight on a project folder
python scripts/check_reproducibility.py .

# Validate all skill files
python scripts/validate_skills.py

Research principles

  • Evidence before rhetoric: connect substantive claims to verifiable sources, data, or reported results.
  • Clarity without distortion: write with confidence while preserving negative findings, contradictory evidence, uncertainty, and material limitations.
  • Reproducibility by design: record search strategies, screening decisions, data provenance, configurations, and hashes.
  • Human accountability: AI-generated suggestions are not peer review, statistical validation, ethics approval, or independent replication.
  • Privacy-aware by default: local tools do not upload manuscripts or datasets to a ReasoningAI server. Network requests occur only when a user explicitly runs a command that requires scholarly metadata.

Network and credentials

OpenAlex and Crossref can be queried without an API key, subject to their current service limits. Semantic Scholar is optional and reads SEMANTIC_SCHOLAR_API_KEY from the process environment. No API keys or personal credentials are included in this repository. See PRIVACY.md and SECURITY.md for data-handling details and limitations.

Verify the release locally

python -m unittest discover -s tests -v
python -m compileall -q reasoningai scripts
python scripts/validate_skills.py
python -m reasoningai secret-scan .

The example bibliographies are synthetic; references-with-duplicates.bib intentionally contains a duplicate DOI to demonstrate an audit finding. The tests use local fixtures and mocks. A passing test suite does not establish that external APIs are currently reachable or that a scientific conclusion is correct. Review generated reports and inspect source records before relying on them.

Contributing

Contributions should improve a real research workflow, preserve evidence integrity, and document the license and provenance of any external material. See CONTRIBUTING.md.

Licensing and attribution

CC BY 4.0 is the default license for this repository, covering all 64 skills, the ReasoningAI Studio and ResearchForge materials, Python utilities and tests, documentation, catalogs, templates, examples, visual assets, and the other project files.

Two prompt files retain their separate MIT license:

  • prompts/en/anti-defensive-revision.md
  • prompts/en/anti-defensive-writing-researchforge.txt

See LICENSE for the repository-wide terms and licenses/MIT-EXCEPTIONS.txt for the exact file-level MIT exception. See NOTICE.md for ownership and provenance details.

About

An evidence-first research toolkit for AI agents and human researchers: 64 focused skills for literature discovery, critical reading, systematic reviews, experiment design, ML evaluation, statistics, reproducibility, scientific writing, and peer review—plus local Python utilities, reusable prompts, and templates for rigorous, transparent research.

Topics

Resources

Contributing

Security policy

Stars

3 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages