English · Italiano
Gives your development agents a memory and the same rules across all projects.
Every agent restarts from what it already knows: context, decisions and tasks. Your data stays on your PC.
flowchart LR
T["You"] --> A["Agent<br/>Claude, Codex, Antigravity"]
A <--> S["SecondBrain<br/>memory, rules, 65 prompts"]
A --> P["Your projects"]
You need Windows, Git 2.42+ and PowerShell 7.5+. First read what gets modified (in Italian).
git clone https://github.com/dennidalpos/SecondBrain.git C:/Progetti/SecondBrain
cd C:/Progetti/SecondBrain
pwsh -NoProfile -File scripts/secondbrain.ps1 -Setup -Interactive
pwsh -NoProfile -File scripts/secondbrain.ps1 -Sync
pwsh -NoProfile -File scripts/mcp-server.ps1 -Pair -Target claude -ApplyThen, in the agent chat, type sb prompt. Step by step: getting-started guide (in Italian).
| Feature | What you get |
|---|---|
| Memory | Context, decisions and tasks of every project, always at hand |
| Common rules | The same instructions for every agent and project |
| 65 ready-made prompts | Development, testing, security, documentation, release; in Italian and English (plus 17 maintenance prompts: sb prompt maintenance) |
| Pre-publish check | Blocks private data and secrets before the push |
| Backup | Every replaced file stays recoverable |
flowchart LR
subgraph PARA["PARA: where things live"]
P1["Projects"] --- P2["Areas"] --- P3["Resources"] --- P4["Archives"]
end
subgraph COSTAR["CO-STAR: how to ask"]
C1["Context"] --- C2["Objective"] --- C3["Style"] --- C4["Tone"] --- C5["Audience"] --- C6["Response"]
end
PARA --> COSTAR --> R["Clear work for the agent"]
- PARA sorts information into four folders. How to use it (in Italian).
- CO-STAR gives every prompt composed by SecondBrain the same structure. How it works (in Italian).
- Progressive Summarization (same author as PARA) names the layered loading: the agent starts from an essential summary and goes down to details only when needed. How it loads (in Italian).
| Tool | Rules file sync | MCP link |
|---|---|---|
| Claude for Windows | supported | -Target claude |
| ChatGPT Codex for Windows | supported | -Target codex |
| Antigravity for Windows | supported | -Target antigravity |
MCP is optional; the same content is also available from the command line.
- With MCP: agent-ready tools and compact answers; at startup and on lookups it picks up queued, already-authorized dossier writes. Deferred writes and permissions (in Italian).
- Without MCP: the agent runs the scripts from a shell; approval requirements depend on the IDE configuration.
Details, diagnostics and limits: agents and MCP (in Italian).
The guides are written in Italian.
| Guide | For |
|---|---|
| Single guide | Everything on one page: installation, IDEs and CLI, variables, concurrency, edge cases |
| Getting started | Install and activate |
| Daily use | Working day to day |
| Per-product flow | The steps in order, in chat and from scripts |
| Prompts | Use and edit the prompts |
| Agents and MCP | Connect the tools |
| Security | Backup, restore, uninstall |
| Technical reference | All the details |
Run these commands from the repository root with PowerShell 7.5+ and Git. Quick is pure (no Git or child processes), with a 15-second budget. The default essential checks have a 90-second budget; Full adds critical boundaries with a 5-minute budget on the reference workstation. For a change without commit or publication, run a final Quick and the relevant sub-suites. Fast runs before a commit; Full runs before publication. Fixtures, logs and a unique run.json report stay in the local runtime after success or failure; -KeepFixtures remains accepted for compatibility. esito.json links the latest report. Reuse evidence only after checking matching inputs and environment; changed inputs invalidate it. See run evidence and limits.
pwsh -NoProfile -File scripts/test-secondbrain.ps1 -Quick -KeepFixtures
pwsh -NoProfile -File scripts/test-secondbrain.ps1 -KeepFixtures
pwsh -NoProfile -File scripts/test-secondbrain.ps1 -Full -KeepFixtures
pwsh -NoProfile -File scripts/build-prompts.ps1 -Check
pwsh -NoProfile -File scripts/check-distribution.ps1Publication also requires SEC07, which checks private data and secrets in the working tree, index and relevant history. Test levels and limits: runtime maintenance.
- Windows only.
- An updated file does not prove that the agent loaded it: check in a new session.
- Tests use sample data, not every real app.
MIT License. Sources and acknowledgements: CREDITS. Central maintenance and backup: runtime.
Dà ai tuoi agenti di sviluppo una memoria e le stesse regole su tutti i progetti.
Ogni agente riparte da quello che sa già: contesto, decisioni e task. I tuoi dati restano sul tuo PC.
Servono Windows, Git 2.42+ e PowerShell 7.5+. Prima leggi cosa viene modificato. I comandi di installazione sono quelli della sezione Get started; poi, nella chat dell'agente, scrivi sb prompt. Passo per passo: guida di avvio.
| Funzione | Cosa ottieni |
|---|---|
| Memoria | Contesto, decisioni e task di ogni progetto, sempre pronti |
| Regole comuni | Le stesse istruzioni per tutti gli agenti e i progetti |
| 65 prompt pronti | Sviluppo, test, sicurezza, documentazione, rilascio; in italiano e in inglese (più 17 di manutenzione: sb prompt maintenance) |
| Controllo prima di pubblicare | Blocca dati privati e segreti prima del push |
| Backup | Ogni file sostituito resta recuperabile |
- PARA ordina le informazioni in quattro cartelle. Come si usa.
- CO-STAR dà a ogni prompt composto da SecondBrain la stessa struttura. Come funziona.
- Progressive Summarization (stesso autore di PARA) dà il nome al caricamento a strati: l'agente parte da un riassunto essenziale e scende ai dettagli solo se servono. Come si carica.
Sync supporta i file delle direttive di Claude per Windows, ChatGPT Codex per Windows e Antigravity per Windows; il caricamento nella sessione va verificato separatamente. Il collegamento MCP è facoltativo (-Target claude, codex o antigravity) e gli stessi contenuti arrivano anche dalla riga di comando.
- Con MCP: strumenti pronti per l'agente e risposte compatte; all'avvio e nelle consultazioni recupera le scritture di dossier già autorizzate e accodate. Scritture differite e permessi.
- Senza MCP: l'agente esegue gli script dalla shell; le richieste di approvazione dipendono dalla configurazione dell'IDE.
Dettagli, diagnosi e limiti: agenti e MCP. Tutte le guide sono nella tabella Guides.
I comandi sono nella sezione Development checks: Quick sulle modifiche, Fast prima del commit e Full prima della pubblicazione, conservando le fixture. Prima del push serve anche SEC07; livelli e limiti nella guida runtime.
- Solo Windows.
- Un file aggiornato non prova che l'agente lo abbia caricato: controlla in una nuova sessione.
- I test usano dati di prova, non ogni app reale.
Licenza MIT. Fonti e riconoscimenti: CREDITS. Manutenzione e backup centrali: runtime.