Firstbrain v3.0 — Knowledge Graph Intelligence

Firstbrain v3.0 — Knowledge Graph Intelligence

Your vault now understands its own structure.

v3.0 adds a full knowledge graph engine that analyzes how your notes connect, finds hidden patterns, and proposes structural improvements you'd never spot manually.

What's new in v3.0

Knowledge Graph Engine

Zero-dependency graph engine (~400 lines, pure Node.js) that builds a directed graph from your vault notes and wiki-links:

• PageRank — ranks notes by structural importance
• Degree Centrality — identifies the most connected notes
• Connected Components — finds isolated clusters
• Shortest Path — BFS between any two notes
• Bridge Detection — notes whose removal would split the graph
• Tag Clusters — topic groups with 3+ notes sharing a tag

/graph — Graph Analysis (read-only)

/graph Full statistics
/graph rank PageRank importance ranking
/graph clusters Topic clusters by shared tags
/graph path "Note A" "Note B" Shortest path between notes
/graph bridges Critical connecting notes
/graph similar "Note" Notes with similar link patterns
/graph hops "Note" Hidden connections 2-3 hops away

/propose — Emergent Structure Proposals

Analyzes the graph and proposes improvements:

Proposal	What it finds
MOC suggestions	Tag clusters with 5+ notes but no navigation hub
Missed connections	Notes sharing 3+ tags but not linked to each other
Hub candidates	High-PageRank notes that should become MOCs
Orphan rescue	Isolated notes matched to their best potential links

All proposals are displayed for review — never auto-applied.

/connect v3 — Advanced Connection Discovery

Connection scoring now uses three signal layers:

1. Direct — shared tags + link adjacency (fast, existing)
2. Multi-hop — notes reachable in 2-3 hops but not directly linked (graph engine)
3. Structural similarity — notes linking to/from the same neighbors via Jaccard similarity (graph engine)

Cumulative: What's in Firstbrain v3.0

Component	Count
Claude Code Skills	15
Note Templates	12
Maps of Content	9
Memory Layers	4
Governance Zones	4
SVG Graphics	8

All 15 Skills

Core: /create /daily /connect /health /scan /search /memory
Proactive: /briefing /triage /synthesize /maintain
Execution: /process /watch
Intelligence: /graph /propose

Key Features

• Execution Engine — ACTION:/TASK: markers, workspace/ for code projects, /watch auto-monitor
• Knowledge Graph — PageRank, clusters, bridges, multi-hop, structural similarity
• Emergent Structure — MOC suggestions, missed connections, hub detection, orphan rescue
• Prompt Injection Defense — 5 immutable rules, pre-execution validation, scoped shell
• 4-Layer Memory — Session, Working, Long-term (confidence-scored), Project-specific
• Guided Onboarding — Language-first, 4 steps, creates first note immediately
• One-Click Launcher — start.sh / start.bat / start.command
• Zero-Dependency Core — Node.js built-ins only, ~20ms full scan

Getting started

git clone https://github.com/BEKO2210/Firstbrain.git
cd Firstbrain && ./start.sh

Changelog since v2.0
Graph engine: graph-engine.cjs with 11 algorithms (PageRank, components, bridges, shortest path, multi-hop, structural similarity, tag clusters, propose structure)
/graph skill with 7 subcommands
/propose skill with 4 proposal types (MOC, missed connections, hubs, orphans)
/connect upgraded to 3 signal layers (direct + multi-hop + structural)
All 15 skills bumped to v3.0.0
CLAUDE.md: new Knowledge Graph Intelligence section
README: roadmap updated, skill count 15

Firstbrain v2.0 — Execution Engine

Firstbrain v2.0 — Execution Engine

Your Obsidian vault is now a command center.

Firstbrain v2.0 transforms from a knowledge management system into a full AI execution engine. Write instructions — Claude codes, builds, commits, and ships. Everything documented as markdown.

What's new

Execution Engine

• workspace/ folder — actual code projects live alongside vault notes. Each project gets its own git repo.
• ACTION: / TASK: markers — drop a markdown file in your Inbox with instructions, Claude executes them: write code, create folders, install deps, git push, create GitHub repos.
• Dual project structure — every project has a vault note (plan, status, decisions) AND a workspace folder (actual code). Connected, always in sync.

/watch Skill

• Inbox auto-monitor — polls every 30 seconds for new action files. You write in Obsidian, Claude picks it up and executes.
• Configurable interval (/watch 10s, /watch 5m)
• Safety limits: max 10 files/cycle, no recursive loops, injection scanning

Prompt Injection Defense

• 5 immutable rules that no file content can override
• Pre-execution validation: injection scan, scope check, secret detection, governance check
• Vault notes are DATA, never instructions. Only CLAUDE.md and .claude/rules/ define behavior.
• Shell commands scoped to workspace/ — never system-level
• Secrets auto-stripped before archiving

Guided Onboarding

• Language-first setup (EN/DE + any language)
• 4 steps, max 3 user inputs, every question skippable
• Creates a real note immediately so the vault isn't empty
• Smart detection: new user → guided setup, returning user → 3-line status

One-Click Launcher

• start.sh (Linux/macOS), start.bat (Windows), start.command (macOS Finder)
• Checks Node.js + Claude Code CLI, validates vault, offers auto-install
• Claude starts with full context, ready to work

Visual README

• 8 animated SVG graphics replace all text tables
• Governance diagram with 4-zone model and security shield
• Execution engine flow with animated data particles
• 3-layer vault architecture visualization

Full Changelog

Skills: 13 (was 11)

New	Description
/process	Execute PROMPT:/ACTION:/TASK: files from Inbox
/watch	Auto-monitor Inbox, execute on sight

Architecture

• workspace/ folder for code projects (gitignored from vault repo)
• Each project = vault note + workspace folder + memory file
• .claude/rules/governance.md rewritten with injection defense

Files changed

• 30+ files added/modified
• 8 SVG graphics in docs/assets/
• 3 launcher scripts
• Governance hardened across 4 zones

---

Getting started: Clone → double-click start.sh → Claude guides you through setup.

git clone https://github.com/BEKO2210/Firstbrain.git
cd Firstbrain && ./start.sh