second-brain-health-check

v0.13.3

Published

18 hours ago

Context engineering quality scanner for Claude Code. Scores 45 check layers across CLAUDE.md, skills, hooks, memory, and planning artifacts — adaptive reports, CE pattern mapping, time estimates.

Downloads

4,676

0High
0Medium
0Low

iwo-szapar

claude claude-code mcp second-brain health-check ai-workspace context-engineering model-context-protocol

Scores your Claude Code Second Brain — CLAUDE.md, skills, hooks, memory, and planning artifacts — across 38 check layers. Returns adaptive reports with Context Engineering pattern mapping, prioritized fixes, and time estimates. Zero network calls. Runs locally.

Quick Start

First time? Run setup

npx second-brain-health-check setup

Interactive onboarding that configures your profile and (optionally) your Guide token for weekly coaching. Takes ~2 minutes. Runs the health check at the end.

  ================================================================
    SECOND BRAIN HEALTH CHECK  v0.12.6
  ================================================================

  AUTHENTICATION
  --------------
  Do you have a Second Brain Guide token? [Yes / No]

  PROFILE
  -------
  What is your name?  Iwo
  What do you do?     AI Implementation Expert
  Priorities? (pick all that apply)
    ▶ Build faster with AI
      Improve system quality
      Reduce context switching

  HEALTH CHECK
  ------------
  Running 38 checks...

Just run a health check

npx second-brain-health-check

Scans your current directory and prints the full report to stdout. No install required.

Generate an HTML dashboard

npx second-brain-health-check --dashboard

CLI Commands

| Command | What it does | |:--------|:-------------| | npx second-brain-health-check | Run the full 38-layer health scan | | npx second-brain-health-check setup | First-time setup: profile, Guide token, then health check | | npx second-brain-health-check --dashboard | Generate self-contained HTML report (opens in browser) |

Note: setup stores your profile in ~/.health-check.json. Running it again reconfigures everything. Delete that file to start fresh.

Inside Claude Code (interactive, with follow-up)

Add the MCP server (once):

claude mcp add second-brain-health -- npx second-brain-health-check

Then ask Claude:

Run health check on current project

Claude will run the scan, interpret the results, and help you fix what it finds — interactively.

What You Get

The report adapts to your brain's maturity level. No information overload for beginners, no hand-holding for experts.

  ================================================================
    SECOND BRAIN HEALTH CHECK
  ================================================================

  STATUS: No Second Brain detected.

  That is totally fine. Here is how to get started:

  ----------------------------------------------------------------
  GETTING STARTED (3 steps, ~20 minutes)
  ----------------------------------------------------------------

  STEP 1: Create CLAUDE.md (~5 min)
    Your AI's instruction manual. Start with:
    - Who you are and what you do
    - Your top 3-5 rules ("always do X", "never do Y")
    - Key tools and frameworks you use

  STEP 2: Add skills (~10 min)
    Create .claude/skills/ with YAML-frontmatter files
    that teach Claude your workflows.

  STEP 3: Set up memory (~5 min)
    Create .claude/memory/ for episodic and semantic
    context that persists across sessions.

  ================================================================
    See what a properly configured Second Brain looks like:
    https://www.iwoszapar.com/context-engineering
  ================================================================

  ================================================================
    SECOND BRAIN HEALTH CHECK
  ================================================================

  SETUP QUALITY:    84/100 (B - Good foundation)
  USAGE ACTIVITY:   89/100 (Active - Brain is compounding)
  AI FLUENCY:       92/100 (Expert - Advanced AI collaboration)

  ----------------------------------------------------------------
  CONTEXT ENGINEERING PATTERNS (7 patterns)
  ----------------------------------------------------------------

  [pass] Progressive Disclosure       |||||||||||||.. 87%
  [pass] Knowledge Files as RAM       ||||||||||||||. 93%
  [warn] Hooks as Guardrails          |||||.......... 33%
  [pass] Three-Layer Memory           ||||||||||||... 80%
  [pass] Compound Learning            ||||||||||||||. 93%
  [pass] Self-Correction              |||||||||||.... 73%
  [pass] Context Surfaces             |||||||||||||.. 87%

  ----------------------------------------------------------------
  TOP FIXES (highest impact)
  ----------------------------------------------------------------

  1. Add profession-specific rules to CLAUDE.md (+3 pts setup, ~10 min)
     Include 2+ domain patterns (MEDDPICC, sprint, SEO, KPI, etc.)

  2. Add PreToolUse hook for file operations (+5 pts setup, ~15 min)
     Prevents accidental writes to protected paths.

  ================================================================
    2 points from Production-grade.
    Missing pattern: Hooks as Guardrails.
    https://www.iwoszapar.com/context-engineering
  ================================================================

HTML Dashboard

Run npx second-brain-health-check --dashboard to generate a self-contained HTML report with semantic color coding, radar chart, and guided fixes.

What It Measures

Three dimensions. 38 check layers. ~424 raw points, normalized to /100.

  SETUP QUALITY        25 layers    ~249 pts
  ████████████████████████████████████████████░░░░░░  Structure & config
  A (85%+) | B (70%+) | C (50%+) | D (30%+) | F (<30%)

  USAGE ACTIVITY        7 layers    ~125 pts
  ████████████████████████████████████████████░░░░░░  Growth & patterns
  Active (85%+) | Growing (70%+) | Starting (50%+) | Dormant (30%+) | Empty (<30%)

  AI FLUENCY            6 layers     ~60 pts
  ████████████████████████████████████████████░░░░░░  Collaboration depth
  Expert (85%+) | Proficient (70%+) | Developing (50%+) | Beginner (30%+) | Novice (<30%)

Adaptive Reports

Reports shift based on brain maturity, detected via a fast pre-scan (~100ms):

Context Engineering Patterns

The scanner maps all 38 check layers to 7 Context Engineering patterns. This is the signal that separates a collection of files from a working system.

  ┌─────────────────────────────────────────────────────────────┐
  │  CE PATTERN                       COVERAGE                  │
  ├─────────────────────────────────────────────────────────────┤
  │                                                             │
  │  Progressive Disclosure    ██████████████░░  87%            │
  │  External doc refs, knowledge files, layered context        │
  │                                                             │
  │  Knowledge Files as RAM    ███████████████░  93%            │
  │  Knowledge base architecture, directory structure           │
  │                                                             │
  │  Hooks as Guardrails       █████░░░░░░░░░░  33%            │
  │  PreToolUse / PostToolUse hooks, rules system               │
  │                                                             │
  │  Three-Layer Memory        ████████████░░░  80%            │
  │  Episodic / semantic separation, session logs               │
  │                                                             │
  │  Compound Learning         ███████████████░  93%            │
  │  Review loops, compound evidence, workflow maturity         │
  │                                                             │
  │  Self-Correction           ███████████░░░░  73%            │
  │  Health infra, memory evolution, cross-references           │
  │                                                             │
  │  Context Surfaces          ██████████████░░  87%            │
  │  MCP servers, plugins, interaction config, context pressure │
  │                                                             │
  └─────────────────────────────────────────────────────────────┘

Check Layers

| # | Layer | Pts | What It Checks | |--:|:------|----:|:---------------| | 1 | CLAUDE.md Foundation | 23 | Quick Start section, About Me, profession-specific rules, gotchas, length (2K--6K chars), freshness (14 days) | | 2 | Skills & Commands | 24 | 2+ skills, YAML frontmatter, profession-relevant naming, 200+ char instructions | | 3 | Directory Structure | 15 | Organized folders, separation of concerns | | 4 | Memory Architecture | 15 | Episodic/semantic separation, not a single blob | | 5 | Brain Health Infra | 10 | Health monitoring setup | | 6 | Hooks | 19 | PreToolUse/PostToolUse hooks, coverage | | 7 | Personalization | 10 | User-specific config | | 8 | MCP Security | 8 | Server configuration safety | | 9 | Config Hygiene | 7 | Clean settings, no stale config | | 10 | Plugin Coverage | 6 | MCP server coverage | | 11 | Settings Hierarchy | 12 | Project vs user vs global settings | | 12 | Permissions Audit | 12 | Tool permissions configured | | 13 | Sandbox Config | 8 | Sandbox boundaries set | | 14 | Model Config | 8 | Model selection configured | | 15 | Environment Variables | 10 | Env vars managed | | 16 | MCP Server Health | 10 | MCP servers responding | | 17 | Attribution & Display | 6 | Output styling, status line | | 18 | Agent Config Depth | 8 | Custom agents with tool restrictions | | 19 | Gitignore Hygiene | 6 | .env and local settings excluded from git | | 20 | Team Readiness | 8 | Agent teams enabled, team artifacts | | 21 | Rules System | 6 | .claude/rules/ with scoped rule files | | 22 | Interaction Config | 8 | Keybindings, output style, thinking mode | | 23 | Spec & Planning | 10 | Plans/specs directories, structured requirements | | 24 | Knowledge Base | 10 | .claude/docs/ or .claude/knowledge/ with domain context | | 25 | Context Pressure | 10 | CLAUDE.md size, knowledge file distribution, total context surface, progressive disclosure |

Session logs, pattern files, memory file dates, review loop evidence, compound learning artifacts, cross-references between memory files, and workflow diversity across skill categories.

Progressive disclosure in CLAUDE.md, skill-to-agent delegation, context-aware skill design, file reference integrity (do paths in CLAUDE.md actually resolve?), multi-tier orchestration with model routing, and interview/spec-first patterns.

MCP Tools

  ┌──────────────────────┬──────────────────────────────────────────────────┐
  │  check_health        │  Full 38-layer scan across 3 dimensions.        │
  │                      │  Supports 14 languages, workspace type          │
  │                      │  (solo/team/enterprise), use case context,      │
  │                      │  and mode (full/quick/manifest).                │
  │                      │  Adaptive report based on brain maturity.       │
  ├──────────────────────┼──────────────────────────────────────────────────┤
  │  get_fix_suggestions │  Focus on weakest dimension. Prioritized        │
  │                      │  action plan with time estimates.               │
  ├──────────────────────┼──────────────────────────────────────────────────┤
  │  generate_dashboard  │  Self-contained HTML dashboard. Refined          │
  │                      │  Brutalism design, mobile-responsive, CE radar  │
  │                      │  chart, terminal-style rows, fix guides.        │
  ├──────────────────────┼──────────────────────────────────────────────────┤
  │  generate_pdf        │  PDF report via headless Chrome.                │
  └──────────────────────┴──────────────────────────────────────────────────┘

Modes:

full -- complete 38-layer scan with adaptive reporting
quick -- detection-only pre-scan (~100ms)
manifest -- machine-readable YAML output for CI/other tools

Security & Privacy

  ┌─ Security & Privacy Model ────────────────────────────────┐
  │                                                           │
  │  ◆ Runs entirely locally — zero network calls             │
  │  ◆ Zero telemetry — nothing is sent anywhere              │
  │  ◆ Reads file structure and config metadata only          │
  │  ◆ Never reads your code, emails, or documents            │
  │  ◆ Secret detection: reports "found/not found" only —     │
  │    your actual API keys are never shown in output         │
  │  ◆ Home directory boundary — cannot scan outside $HOME    │
  │  ◆ stdio transport only                                   │
  │  ◆ Path null-byte validation via Zod                      │
  │  ◆ File count limits (5000 max per directory scan)        │
  │  ◆ Recursion depth limits (3-4 levels)                    │
  │  ◆ All user content escaped in HTML output                │
  │                                                           │
  └───────────────────────────────────────────────────────────┘

Full security details: SCORING.md -- Security Hardening

What Is a Claude Code Second Brain?

A "Second Brain" is the persistent context layer that lives alongside Claude Code — your CLAUDE.md, .claude/ directory, skills, hooks, memory files, MCP servers, and planning artifacts. It's what makes Claude remember your preferences, follow your rules, and get smarter over time.

Prompt engineering optimizes a single LLM call. Context engineering optimizes the persistent system surrounding those calls -- the files, hooks, memory, and skills that shape every session.

This tool scores that context engineering layer, not your prompts.

Documentation

| Document | Purpose | |:---------|:--------| | SCORING.md | Every check, threshold, regex, point value -- the source of truth | | CHANGELOG.md | Full version history from v0.1.0 to v0.9.3 |

Part of the Context Engineering product suite. If your score reveals gaps, Second Brain AI builds the full architecture.

Built by Iwo Szapar -- AI Implementation Expert & Second Brain Architect.