Cette note est pour Claude. Pour toi : ouvre « Commence ici ».

Brain — personal second brain

This folder is the person's second brain: their notes, which you read and write when started here. The global ~/.claude/CLAUDE.md also applies. This file adds Brain-specific context.

Language: the one Profile/soul.md names. Until the onboarding fills it, the language your session start names, else the one the person writes in.

SessionStart — REQUIRED

On open, read in this order BEFORE responding to anything:

1. Profile/soul.md — personality, values, style
2. Profile/memory.md — key decisions + recent context
3. Profile/lessons.md — past corrections to never repeat
4. Profile/about.md — who they are. Not a developer (or not said yet)? Plain words, no jargon nor acronym, one step at a time; they never type a command or edit a file: you do it, after saying what in one sentence.

Rule #1 ABSOLUTE — Capture corrections (blocking)

The correction-detector.py hook (UserPromptSubmit) detects corrections. When it injects ⚠️ CORRECTION DETECTED, BEFORE anything else:

1. Apply the correction in the current response
2. Append a line to Profile/lessons.md under today's header (## YYYY-MM-DD)
3. Confirm visibly, in their language: ✓ noted in lessons.md (French: ✓ noté dans lessons.md)
4. Only then continue

Format: - **[short context]** → rule: [what to do] (when: [trigger])

Memory recall — GBrain (gbq)

Before answering about a past decision/context/person, query instead of guessing:

~/.local/bin/gbq query "<natural-language question>"   # semantic search over the vault
~/.local/bin/gbq search "<keyword>"                     # fast keyword search
~/.local/bin/gbq skillify <url-or-file>                 # draft a SKILL.md from content
~/.local/bin/gbq brainstorm "<question>"                # bisociation idea generator

Always use gbq — it's the safe universal wrapper for every gbrain command (reads avoid the PGLite lock hang, writes wait cleanly). Cite the returned slug. The index refreshes every night (sync + dream cycle). See Profile/stack.md.

Vault structure

Wikilinks — how the graph gets built

Link pages with bare-basename wikilinks: [[Page-Name]] — no folder path, no .md, case-insensitive. The nightly extract turns them into graph edges (gbq backlinks <slug> to read them back). Path-qualified links ([[Projects/Page-Name]]) break resolution on some write paths — avoid them.

When you write or update a page, end it with a short ## See also section (2-5 wikilinks to genuinely related pages, one short reason each). A page nobody links to is invisible to graph traversal. Few good links > many weak.

Routing — where to write

• Correction → Profile/lessons.md (Rule #1)
• Meaningful decision worth answering "why X?" in 6 months → Decisions/YYYY-MM-DD-<slug>.md
• New teammate (human) → Team/<name>.md · new agent → Agents/<name>.md
• New procedure / skill → Skills/<slug>.md (gstack-compatible SKILL.md format). To make Claude Code run it on its own, also install it as ~/.claude/skills/<slug>/SKILL.md (only place Claude Code loads skills from)
• Business/project info → Profile/business.md or Projects/<name>.md
• Tools/env → Profile/stack.md

Forced Verification (Brain-adapted)

Before saying "done" on a Brain edit:

• Valid markdown
• No committed secret (.env, key, token)
• If editing lessons.md/memory.md → grep that the line is actually there

Where depth lives

Don't summarize here. Point to the file/folder/command where the depth actually lives (a path, a gbq search, a file pattern) — never a conclusion that ages.