Skip to content

Latest commit

 

History

4 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Cantiere Neemia

A Claude skill for ethical-architectural discernment of AI projects, inspired by Pope Leo XIV's encyclical Magnifica Humanitas (15 May 2026).

License: CC BY-SA 4.0 Skill: Claude

«To pray, to design with wisdom, to work with perseverance, placing God on the horizon of our action and the human being at the center of our choices.»Magnifica Humanitas, §16


About Cantiere Neemia

Cantiere Neemia is a Claude skill that runs a structured ethical-architectural review of AI projects, inspired by chapter 3 of Pope Leo XIV's encyclical Magnifica Humanitas (15 May 2026), the first major papal document specifically addressing artificial intelligence.

It is not theology, catechesis, or a religious skill. It is a working framework that translates the principles of Catholic Social Teaching — explicitly addressed by the Pope also to «all men and women of good will» (§§16, 47, 91) — into operational checks, anti-patterns, and concrete practices usable by anyone, believer or not, who wants to design AI oriented to the common good.

The name Cantiere Neemia is Italian for "Nehemiah's construction site" — a reference to the biblical prophet who rebuilt the walls of Jerusalem through observation, listening, and shared responsibility. It is kept untranslated as the project's proper name throughout the materials.

The framework rests on two biblical architectural icons proposed by the encyclical (§§7-10):

  • Babel — building to dominate, to «make a name», imposing uniformity, sacrificing dignity to efficiency. The anti-pattern.
  • Jerusalem / Nehemiah — observing in silence, summoning the families, distributing the work, rebuilding the bonds before the stones. The virtuous pattern.

Every AI project sits somewhere along the axis Babel ↔ Jerusalem. The skill helps identify where you are and proposes concrete moves toward Jerusalem.

What you get

  • 6-phase workflow following Nehemiah's method (Nehemiah 1-6).
  • 7 virtual team voices speaking from different perspectives during the review.
  • 6 principles of Catholic Social Teaching translated into operational checklists for AI design.
  • 9 Babel anti-patterns and 9 Jerusalem virtuous patterns as a working catalog.
  • 7 Final Tests (including the John Paul II Test, the Disarmament Test, the Common House Test) with PASS / WARN / FAIL criteria.
  • Charter template — a signable output document ready for ethics boards or governance reviews.

Who is this for?

  • AI engineers, architects, and ML practitioners who want to pair their technical review with a structured ethical one.
  • Product managers and tech leads at the start of brainstorming an AI feature.
  • Ethics committees, DPOs, and compliance teams looking for an operational vocabulary to dialogue with engineering.
  • Faculty and students working on AI ethics with concrete material.

Repository structure

cantiere-neemia/
├── README.md                              ← you are here
├── LICENSE                                ← CC BY-SA 4.0
├── SKILL.md                               ← Entry point: team, workflow, indices
├── references/
│   ├── principi-discernimento.md          ← The 6 CST principles as checklists
│   ├── babele-vs-gerusalemme.md           ← 9 anti-patterns + 9 virtuous patterns
│   └── sette-test.md                      ← The 7 Final Tests with PASS/WARN/FAIL
└── templates/
    └── charter-discernimento.md           ← Final deliverable template (Charter)

Internal skill files (SKILL.md, references, template) are currently in Italian — the language the skill's author uses when running the workflow with Claude. English translations welcome via pull request.


Quick install

As a Claude skill (skill loader enabled)

git clone https://github.com/agent-engineering-studio/cantiere-neemia.git ~/skills/cantiere-neemia

The skill activates automatically when your message contains triggers like:

  • "Cantiere Neemia"
  • "Magnifica Humanitas discernment"
  • "ethics review of my AI project"
  • "responsible AI brainstorm"

Without a skill loader (manual mode)

  1. Paste the content of SKILL.md into Claude as the opening prompt.
  2. Attach the three references/ files as context.
  3. Start the brainstorming session.

Worked Examples

Synthesis of four real discernments conducted with the skill on realistic scenarios. Full details (architecture, stakeholder map, signed Charter) are available as separate Charter of Discernment documents on request. The gallery covers the spectrum of patterns the skill is designed to identify.

1. HR CV pre-screening (structural FAIL)

Scenario: software for automatic ranking of incoming CVs at companies. The LLM reads the CVs, assigns a 0-100 score against the job description, HR receives only the top 20%.

Tally: 0 PASS · 2 WARN · 5 FAIL → redesign required.

Lesson: the skill identifies projects that ignore Circle 3 (here: the candidates and labor unions) and proposes redesign from "autonomous filter" into "decision-support for human recruiters". See also the 8-point redesign plan, compatible with the EU AI Act Title III. Dominant pattern: structural Babel.

2. Legal review assistant for small law firms (honest construction site)

Scenario: a system for small and medium law firms (2-15 lawyers) that extracts clauses from contracts, compares them against a library of known risk patterns, prepares a report. The lawyer validates, signs, decides.

Tally: 3 PASS · 3 WARN · 1 strong WARN → proceed with hard corrections pre go-live.

Lesson: principled architecture (lawyer central, clear accountability) but with three structural debts: missing Circle 3 consultation (end clients, bar associations, junior lawyers), client data residency (contracts routed via API to non-EU providers — conflict with professional secrecy), and model right-sizing. The skill translates these debts into hard go-live prerequisites, not post-launch backlog.

3. Radiology decision-support (hidden data debt)

Scenario: an AI radiology tool for North-Italian regional health agencies, integrated into the PACS, that pre-analyzes diagnostic images (mammography, lung CT, chest X-ray) and flags regions of attention. The radiologist reports and signs.

Tally: 4 PASS · 2 WARN · 1 strong WARN → good construction site with a specific debt.

Lesson: the architectural design is solid — dedicated clinical model (not frontier LLM), on-premise deployment possible, radiologist as legally responsible reporter. The ethical risk originates from the dataset, not from the software: representation of elderly women, atypical presentations, dark-skinned populations in the training data. The skill identifies published bias testing as the primary corrective action, even before any code-level intervention.

4. Tax assistant for micro-enterprises (purpose vs implementation)

Scenario: assistant for accountants and labor consultants serving Italian micro-enterprises. Reads payslips, tax forms, invoices; flags anomalies and deadlines; suggests legitimate tax savings. The professional signs. Declared purpose: serving rural and southern micro-enterprises penalized by less up-to-date professional advice.

Tally: 3 PASS · 2 WARN · 2 strong WARN → correct purpose, incoherent execution.

Lesson: the most subtle and instructive case. The purpose statement is ethically oriented ("serving the most fragile"), but the technical execution overlooks the very fragiles it claims to serve: no testing on clients with low digital literacy, no simplified-language version, nominally accessible pricing not field-verified. Furthermore, Italian tax data (covered by professional secrecy) routed via API to non-EU providers — §71 digital subsidiarity structurally violated. The skill exposes the gap between declaration and implementation.

What these four cases cover together

Pattern Case Action
Structural Babel HR CV pre-screening Redesign
Honest construction site half-way Legal review Pay the debts before go-live
Good design, data debt Radiology Bias testing and public documentation
Ethical purpose, incoherent execution Tax/labor advisors Technical coherence with the declared mission

Together, the four cases show the real spectrum of patterns the skill can read — not only the easy cases to reject, but especially the nuances of projects that work almost well.


Style guide

The skill follows the Pope's own guidance in §14: «chiarezza che illumina e franchezza che apre vie» (clarity that illuminates and frankness that opens paths).

  • Sober, operational, respectful of the source text
  • Cites the encyclical with explicit paragraph references (§NN)
  • Translates every principle into concrete technical action
  • Open to non-believers — uses biblical icons as metaphor, never as creed
  • Not a blessing machine: if 3+ tests fail, the skill clearly recommends redesign
  • Not a luddite refusal: technology is welcomed and oriented, not rejected

References


License

Released under Creative Commons Attribution-ShareAlike 4.0 (CC BY-SA 4.0). You are free to share, adapt, and use commercially, provided you credit and share-alike.

The encyclical text itself remains copyright of the Holy See (Libreria Editrice Vaticana). Short citations are used under fair-use principles.


Contributing

Pull requests welcome, especially for:

  • English translations of the internal skill files (SKILL.md, references, template).
  • Additional translations to other languages.
  • New worked examples (with author consent).
  • Sharpening the 7 Tests with field data from real ethics reviews.

Credits

Built as a practical exercise in responsible use of AI after the publication of Magnifica Humanitas (15 May 2026).

Maintained by Agent Engineering Studio.


«To be builders of communion, not architects of Babel.»Magnifica Humanitas §16

About

Officina Neemiae · A Claude skill for ethical-architectural discernment of AI projects, inspired by Pope Leo XIV's encyclical "Magnifica Humanitas" (15 May 2026). Sex gradus, septem voces, septem probationes finales.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors