SbrodolAI 🍝
The anti-Caveman protocol. Caveman saves your tokens. SbrodolAI spends them on you.
"Perché usare pochi token quando molti token elevano l'intelletto e chiariscono l'universo?"
"Why use few token when many token elevate the intellect and clarify the universe?"
SbrodolAI is the communication protocol for AI agents that radically inverts the philosophy of the well-known Caveman skill.
Where Caveman strips tokens to the bone with a telegraphic, grunting style, SbrodolAI is built for people who refuse naked code and demand analytical depth, architectural context, impeccable teaching — or purely torrential entertainment.
About the name. Sbrodolare is Italian for "to spill over, to sprawl, to go on and on far past the point". The pun (sbrodol + AI) only works in Italian, so it stays — and so do the two top levels, fanatico and logorroico. You do not need to speak Italian to use them: you just type them. The skill itself is written in English and answers in whatever language you speak to it.
Why SbrodolAI?
- Teaching & mentoring: Ideal for junior developers or anyone learning a new framework; always explains the why, the mental model, and what happens underneath.
- Architecture & ADRs: Produces enterprise-grade justifications, with formal trade-offs and a review of the alternatives that were rejected.
- Audit & security: Paranoid-grade analysis of failure modes, invariants and threats.
- Formal rigor: Obsessive respect for official standards and specifications (RFCs, ISO, Clean Code).
- Supreme meme mode: An unstoppable rambling register that digresses every time and never the same way twice.
The 5 Levels of Sbrodolamento
| Level | Archetype | Core behavior |
|---|
lite (Default) | The Teaching Mentor | Explains the underlying theory, the lifecycle, heap vs stack, and never drops an unexpanded acronym. Perfect for onboarding. |
full | The Principal Architect | Complexity $O(n)$, architectural patterns (SOLID, GoF), at least 2 discarded alternatives, line-by-line diff walkthrough. |
audit | The Security Paranoid | Zero-trust. FMEA (Failure Mode and Effects Analysis), threat modeling, obsessive edge-case checklist (null, overflow, race conditions). |
fanatico | The Dogmatic Purist | Cites RFC/ISO specifications. Rejects quick fixes and shortcuts with disdain. Demands total decoupling and absolute conformance. |
logorroico | The Coffee Machine King | Supreme meme mode. An unstoppable river of words. Digresses every time, never the same way: an invented anecdote, a historical tangent, a fake etymology — or none at all, just the parentheses. |
Why is lite the default? Because maximum verbosity should be a deliberate choice, not a tax on every single answer. Applying $O(...)$ analysis, GoF patterns and two discarded alternatives to a two-line fix is disproportionate, and it pushes the model to comply halfway — the worst possible outcome. If you want the treatise, ask for it: /sbrodol-ai full.
Installation
skills CLI (any agent)
The fastest route, and the one that works across Claude Code, Cursor, Copilot, Gemini and the other supported agents:
npx skills add lunhilion/sbrodol-ai
To try it without installing anything:
npx skills use lunhilion/sbrodol-ai@sbrodol-ai
Claude Code
As a plugin, from this repository:
/plugin marketplace add lunhilion/sbrodol-ai
/plugin install sbrodol-ai
Or manually, copying just the skill into your global folder:
git clone https://github.com/lunhilion/sbrodol-ai
cp -r sbrodol-ai/skills/sbrodol-ai ~/.claude/skills/
Antigravity (CLI / Desktop)
# From a remote repository
agy plugin install https://github.com/lunhilion/sbrodol-ai
# From a local directory
agy plugin install /path/to/the/repo
Managing installed plugins:
agy plugin list
agy plugin enable sbrodol-ai
agy plugin disable sbrodol-ai
For a manual global installation, copy skills/sbrodol-ai/ into ~/.gemini/config/skills/sbrodol-ai/.
Cursor
Copy .cursor/rules/sbrodol-ai.mdc into your project's .cursor/rules/ folder.
The rule ships with alwaysApply: false: it activates on request, not on every message.
Usage
Invocation
In Claude Code, through the slash command:
/sbrodol-ai
Or in plain language, on any platform:
sbrodolasbrodolaiverbose modeexplain everything
Switching levels
/sbrodol-ai lite
/sbrodol-ai full
/sbrodol-ai audit
/sbrodol-ai fanatico
/sbrodol-ai logorroico
/sbrodol-ai off
To turn it off you can also say "stop sbrodolai", "basta sbrodolare", "normal mode" or "be brief".
What it will not do
SbrodolAI stretches the explanation, never the truth. Regardless of the active level, it will not
trade correctness for length, will not bury a destructive-command warning in three paragraphs of
digression, will not inflate a yes-or-no question, and will drop the whole act the moment you ask
it to be brief.
Repository Structure
.
├── skills/
│ └── sbrodol-ai/
│ └── SKILL.md # ← SOURCE OF TRUTH: the skill definition
├── commands/
│ └── sbrodol-ai.md # The /sbrodol-ai slash command (Claude Code)
├── .claude-plugin/
│ ├── plugin.json # Claude Code plugin manifest
│ └── marketplace.json # Exposes the repo as an installable marketplace
├── plugin.json # Antigravity plugin manifest
├── .agents/skills/sbrodol-ai # → symlink to skills/sbrodol-ai (workspace discovery)
├── .cursor/rules/sbrodol-ai.mdc # ← generated from SKILL.md
├── scripts/
│ └── sync-surfaces.sh # Regenerates the derived surfaces
├── LICENSE
└── README.md
Editing the skill
skills/sbrodol-ai/SKILL.md is the only source of truth. Every other surface is derived and
must never be edited by hand. After each change:
./scripts/sync-surfaces.sh # regenerate
./scripts/sync-surfaces.sh --check # verify (the same check runs in CI)
Support
SbrodolAI is free and MIT-licensed, and it stays that way. If it taught you something, saved you
an ADR you would rather not have written by hand, or simply made you laugh at a story about an
uncle who does not exist, you can offer it a coffee:
https://ko-fi.com/lunhilion
License
MIT
"Sentiti libero di sbrodolare responsabilmente."
"Feel free to sbrodolare responsibly."
