Humanities Superpowers

한국어 · Documentation · Quick start · White paper · Manifesto · Examples

Structured research skills and quality gates for humanities scholars using AI agents.

Use AI without surrendering scholarly judgment.
Not an AI paper writer. A scholarly judgment scaffold.

Humanities Superpowers helps researchers slow down the moments where fluent AI output is most dangerous: framing a question, defining concepts, connecting claims to evidence, testing objections, checking citations, and deciding whether a manuscript is ready to submit.

It provides 13 core research skills and one Level 3 router that selects the smallest valid route through them. It supports scholarly judgment; it does not replace interpretation, source verification, ethics, or authorship, and it does not promise a publication-ready manuscript.

Critical testing invited

The initial goal is to recruit 10 humanities researchers willing to test one research skill on a real project and report where the framework is too rigid, too technical, or methodologically weak.

• Submit a test report
• Offer methodological criticism
• Report an installation problem
• Join a broader discussion

Literary scholars, historians, philosophers, digital humanists, DH librarians, graduate researchers, and scholars working in languages other than English are especially welcome. Do not post unpublished manuscripts, personal data, reviewer identities, or copyrighted source text; use anonymized descriptions and minimal excerpts.

Quick start

git clone https://github.com/icerain-cmd/humanities-superpowers.git
cd humanities-superpowers
python3 scripts/validate_repository.py

On Windows, use python instead of python3 if that is the available launcher.

Copy skills/ into your agent's skill directory, or follow INSTALLATION.md.

Then ask:

Use Humanities Superpowers to formulate a research question about artificial nature and platform aesthetics. Separate verified evidence, interpretation, inference, hypothesis, and unknowns. Do not invent sources.

For a complete workflow:

Follow workflows/write-a-paper.md. Stop at every quality gate and report unresolved risks before proceeding.

Verification and compatibility

Tested: Claude Code and OpenAI Codex.

Installation guidance provided, but not yet independently verified: Cursor.

Other Markdown-capable agents may use the skills manually, but compatibility is not guaranteed. Product conventions can change; see INSTALLATION.md for the tested project layouts, file-count checks, and a read-only pilot workflow.

Why it exists

AI agents can produce polished academic prose before the underlying research has been verified. Humanities research requires a different discipline: conceptual lineage, close reading, argumentative restraint, traceable evidence, serious counterarguments, and explicit uncertainty.

The framework therefore organizes work around three commitments:

• Research before prose
• Evidence over confidence
• Completion must be demonstrated

Read the methodology white paper, project philosophy, and design principles.

13 core research skills + 1 Level 3 router

Each skill defines invocation conditions, required inputs, a procedure, stop signals, anti-fabrication rules, completion criteria, output records, and next steps.

The separate using-humanities-superpowers router diagnoses research state and selects among these 13 core skills. The repository therefore contains 14 SKILL.md files: 13 core research skills and 1 router.

Research quality gates

A gate returns:

• PASS — no known blocking issue remains;
• CONDITIONAL PASS — recorded risks are explicitly accepted;
• FAIL — evidence, verification, or researcher judgment is still missing.

A failed gate is not a broken workflow. It is a refusal to hide unresolved scholarly risk. See QUALITY_GATES.md.

Human–AI division of responsibility

AI may organize, compare, test, and flag. The researcher remains responsible for significance, interpretation, source verification, ethics, conceptual commitments, disclosure, and final authorship.

Worked examples

End-to-end concept paper

Artificial Nature and Platform Mediation runs from a research brief through question formation, scope, concept lineage, argument mapping, terminology control, citation audit, and final verification.

The final gate intentionally returns FAIL. No verified source set is supplied, so the framework refuses to label the project publication-ready. This is expected behavior.

Focused examples

• Argument-map example
• Terminology-audit example

What this project does not claim

Humanities Superpowers does not guarantee truth, originality, acceptance, or citation accuracy. It does not turn an AI agent into an autonomous scholar. It reduces avoidable risk by making assumptions, evidence, unresolved verification, and researcher decisions visible.

See ANTI_PATTERNS.md for common failure modes.

Origin and independence

Humanities Superpowers was inspired by Jesse Vincent's obra/superpowers, which applies composable skills and systematic verification to coding agents. This project independently adapts that general design idea to humanities research.

It is not affiliated with, endorsed by, or maintained by Jesse Vincent, Prime Radiant, or the Superpowers project. See ACKNOWLEDGMENTS.md, DIFFERENCES.md, and THIRD_PARTY_NOTICES.md.

Author

Lee Yong Wook · Jeonju University
GitHub: @icerain-cmd · [email protected]

Contributing, citation, and license

Read CONTRIBUTING.md before opening an issue or pull request. Citation metadata is provided in CITATION.cff. Released under the MIT License.

Research orchestration

The Level 3 router diagnoses the current research state, selects the smallest valid skill route, and treats failed gates as reasons to pause or roll back—not as obstacles to hide. It can preserve a resumable session record without treating memory as evidence.

• Research state machine
• Routing and rollback
• Session memory

Documentation site

The live documentation is published at icerain-cmd.github.io/humanities-superpowers. It is built with MkDocs Material and deployed from main through GitHub Actions.

Public release status

The v1.0.0 release notes and final publication checklist are available in docs/release/.

Social media launch kit

• Social media launch kit