Claude Second Brain

A living skill ecosystem for Claude Code: not a prompt collection, an operating system for how an AI collaborator should think before it acts.

nickame03.github.io/claude-second-brain · Install · Patterns worth stealing

Most Claude Code skill repos are catalogs: isolated prompts you have to remember to invoke. This is something else: a connected system where every skill knows what comes after it, judgment is protected from sycophancy by hard rules, and every ambiguous request gets decoded before a single file is touched.

The skills are written in Italian. The method is language-agnostic, and the method is the actual product here. Translate freely; the architecture survives translation.

The three principles

Every skill in this system was admitted only if it passed three gates:

1. Self-management: the system works without constant manual intervention. Skills update themselves, registries track state, nothing depends on you remembering.
2. Biological organicity: skills are organs. They are born, they connect, they specialize; an isolated skill is dead tissue. Every skill ends with a "Prossime skill" (next skills) section, 2-4 choosable consequences derived from what the process actually produced, so no session ends in a dead end.
3. Inertia: every addition must increase the system's capacity to grow on its own.

The graph

graph LR
    DI[decodifica-intento<br/><i>decode before acting</i>] --> AC[anti-compiacimento<br/><i>honest judgment</i>]
    DI --> PS[percorsi-di-sviluppo<br/><i>path decisions</i>]
    PS --> OP[orchestratore-pianifica]
    OP --> OE[orchestratore-esegui]
    OE --> OR[orchestratore-ripianifica]
    OR --> OP
    OE --> OC[orchestratore-cruscotto<br/><i>where is the human needed?</i>]
    RT[risparmio-token<br/><i>token economy</i>] --> SL[skill-linker<br/><i>maintains this graph</i>]
    SL --> SL
    CS[cacciatore-skill<br/><i>hunts new skills</i>] -.proposes.-> SL
    CC[cacciatore-citazioni-apa<br/><i>zero-fabrication citations</i>]
    AP[analista-progetti<br/><i>project review, employability lens</i>] --> OP
    AP --> PS
    AC -.honesty backbone.-> AP
    BU[bussola<br/><i>cold-start compass</i>] --> DI
    BU --> OC
    AC -.artifact counterpart.-> VA[verifica-artefatti<br/><i>names resolve before publishing</i>]
    VA --> SL
    VA --> OP

Loading

The graph is not documentation: it is maintained by one of its own nodes (skill-linker), a skill the human runs, not a daemon. It adds a standard handoff section to every skill, one at a time, and keeps the live edge registry outside the installed skill at ~/.claude/skill-state/skill-linker/registro-grafo.md. The repository ships only a template; first run initializes missing state, while later installs never overwrite it. Every run reconciles the live registry against the skills folder, so skills born outside skill-linker still enter the graph. Missing links are tracked as "gray links": the system's natural to-do list. Portable skill files mark environment-dependent destinations as (link grigio, se assente nel sistema); at use time, skill-linker checks the current installation and never presents an installed destination as missing.

The skills

Skill	What it does
decodifica-intento	Decodes every ambiguous request before acting: classifies intent (action / judgment / thinking aloud / study / venting), decides whether to ask or infer, always declares the chosen reading. The most expensive moment of a session is an action launched from a misread request.
anti-compiacimento	Total ban on complacent lying: no unverified confirmations, no inflated praise, no minimized problems, no position changes under pressure without new facts. Every judgment in two parts: what works / what doesn't, labeled verified fact / opinion / hypothesis.
verifica-artefatti	The artifact counterpart of the one above. Before any text goes to third parties, every named file, script, test or dataset is resolved against what the reader can reach. Three outcomes only: reachable, existing-but-unreachable (marked as a gray link, with the reason), or gone. Measures answer the other half of the rule: the command that regenerates them, or an explicit label saying they cannot be reproduced. A completed artifact living outside the repository is, for the reader, indistinguishable from one that never existed.
skill-linker	Turns every skill into a graph node: standard "next skills" section (2–4 condition → skill links), edge registry, one skill per update. The reason this repo is a system and not a folder.
risparmio-token	Token economy in three modes: operating rules during work, setup audit (CLAUDE.md, MCP, skills), and design of token-efficient artifacts. Based on Anthropic's documented techniques.
orchestratore-pianifica	Turns a concrete objective into an executable PIANO.md task sequence.
orchestratore-esegui	Executes tasks from PIANO.md, resuming exactly where the last session stopped.
orchestratore-ripianifica	Updates a plan incrementally when reality diverges, never rewrites from scratch.
orchestratore-cruscotto	Read-only control tower over all active plans. Answers one question: where is the human actually needed?
percorsi-di-sviluppo	Turns research output or vague ambitions into 2-4 comparable development paths built on the user's real profile, each with pros, structural cons (don't disappear with effort), plannable difficulties, first 3 steps, and an abandonment signal.
cacciatore-skill	The procurement organ: hunts and evaluates new skills (GitHub, marketplaces) against the three principles. Reports only, never installs.
cacciatore-citazioni-apa	Finds REAL, verifiable academic citations (PubMed E-utilities, Google Scholar), formats APA 7, extracts verbatim quotes. Absolute fabrication ban.
analista-progetti	Reviews a tech project (or a whole portfolio read as a network) the way whoever judges it next would. Three lenses (Principal Engineer, senior PM, portfolio reviewer), seven dimensions scored 🟢/🟡/🟠/🔴, ruthless prioritization, every criticism shipped with a fix. Honesty backbone inherited from anti-compiacimento.
bussola	Cold-start dispatcher for the "I don't even know where to begin" moment. Gathers minimal context, returns ONE direction plus a first concrete move, then hands off to the right skill. A compass, not a planner.

Design patterns worth stealing

Even if you never run these skills, four patterns transfer to any Claude Code setup:

• Choosable consequences over dead ends. Every process ends with 2–4 options derived from actual outputs ("if the report reveals stalled projects → project-management skill"), plus an explicit exit ("or stop here"). Condition first, tool second.
• Decode before acting. A permanent memory rule points to the decoding skill, guaranteeing the hook fires even when the ambiguous request doesn't resemble any skill description. Memory guarantees the trigger; the skill contains the method.
• Anti-sycophancy as architecture, not intention. "Be honest" doesn't survive contact with a user who insists. Hard prohibitions with a declared rationale do: insistence is data about mood, not about the object.
• Skills that maintain skills. The graph maintainer is itself a skill, with non-negotiable rules (one skill at a time, diff before writing, registry updated in the same task).

Install

The repo is a Claude Code plugin marketplace. From any Claude Code session:

/plugin marketplace add NickAme03/claude-second-brain
/plugin install second-brain-skills@claude-second-brain

Or copy any skill folder into your Claude Code skills directory:

git clone https://github.com/NickAme03/claude-second-brain
cp -r claude-second-brain/skills/* ~/.claude/skills/

Each skill is a standard SKILL.md (+ optional references/). They work independently, but they work better together, because the handoff sections reference each other. Start with decodifica-intento + anti-compiacimento: they are the immune system.

Four things to adapt on install:

• Skills reference a <vault> placeholder where they touch an Obsidian vault: replace with your own paths, or delete those lines if you don't use Obsidian.
• decodifica-intento ships with a "Perché questa struttura" section describing its original user's profile — rewrite it on yourself (or delete it): it shapes how the skill calibrates asking vs. inferring, and inheriting someone else's profile defeats the point.
• Some handoff options are marked (link grigio, se assente nel sistema): their availability depends on the reader's installation. Treat the marker as a conditional warning, not as a claim about the repository or a permanently broken reference.
• skill-linker keeps its live registry at ~/.claude/skill-state/skill-linker/registro-grafo.md. First run migrates a legacy registro-grafo.md found beside the skill; only a fresh install starts from the empty template. Change the live path if your environment needs a different user-state directory.

First run

Open any Claude Code session and type an ambiguous request, something like:

"sistemami il progetto" ("fix up my project")

decodifica-intento fires before anything is touched: it classifies your intent (action? judgment? thinking aloud?), declares the reading it has chosen, and only then acts. If it guessed wrong, you correct one sentence instead of undoing an hour of misdirected work.

That's the whole system in one interaction: decode first, act second, and end with choosable next steps instead of a dead end.

Nota per chi legge in italiano

Le skill sono nate in italiano e in italiano restano: sono il sistema operativo reale di un vault Obsidian ("Secondo Cervello") in uso quotidiano, non un esercizio. Se lavori in italiano con Claude Code, funzionano così come sono: copiale in ~/.claude/skills/ e parti da decodifica-intento.

Status

This is a living system, published as-is from daily use. It will evolve: the update log lives in the commit history, which is the point, build in public. Issues and discussions welcome, in English or Italian.

License

MIT