BoardClaude

Assemble a board of AI agents that evaluate your project from multiple expert perspectives.

BoardClaude is a Claude Code plugin that creates configurable panels of AI evaluator agents. Each agent brings a distinct perspective -- architecture, product, innovation, code quality, documentation, community impact -- and scores your project against weighted criteria. Then it fixes the issues it finds, validates the fixes, and re-audits to prove improvement.

Features

• Multi-Agent Evaluation -- 6 specialized agents score your project in parallel using Agent Teams
• Closed-Loop Improvement -- Audit, fix, validate, re-audit. Track score progression across iterations
• Configurable Panels -- YAML-defined agent panels. Use built-in templates or create your own
• Real Validation -- Agents cite real tsc, jest, eslint, prettier output, not opinions
• Visual Dashboard -- Web UI with radar charts, agent cards, score progression, timeline visualization
• Panel Templates -- YAML-defined agent panels with 4 built-in templates, or create your own

Prerequisites

• Claude Code CLI installed and authenticated
• Node.js 20+ (for the web dashboard)
• Git

Quick Start

Install

# Option 1: npx (recommended)
npx boardclaude install .

# Option 2: Global install
npm install -g boardclaude
boardclaude install .

# Option 3: One-command install from source
git clone https://github.com/ojallington/boardclaude.git
./boardclaude/install.sh your-project/

Run Your First Audit

# In any project with the plugin installed
/bc:init           # Setup wizard - choose a panel template
/bc:audit          # Run full panel audit (6 agents + synthesis)
/bc:fix            # Implement top action items from audit
/bc:audit          # Re-audit to measure improvement

Expected output after a successful audit:

Audit complete: audit-20260210-193000
  Composite: 68.4 / 100 (C+, MARGINAL)
  Agents: boris=69.0, cat=71.2, thariq=71.0, lydia=68.1, ado=64.2, jason=62.4
  Action items: 10 (5 low, 3 medium, 2 high)
  Saved to: .boardclaude/audits/audit-20260210-193000.json

Available Commands

| Command | Description | |---------|-------------| | /bc:audit [target] | Run full panel audit with all agents | | /bc:init | Setup wizard, choose template or create custom panel | | /bc:fix [--max N] | Implement top-N audit action items, validate, re-audit | | /bc:review [agent] [target] | Quick single-agent review | | /bc:fork [name] | Create strategy branch via git worktree | | /bc:compare [a] [b] | Side-by-side branch or audit comparison | | /bc:merge [branch] | Integrate winning branch, archive losers |

Web Dashboard

cd dashboard
npm install
npm run dev
# Open http://localhost:3000

Screenshots

| Landing Page | Audit Results | Audit Detail | |:---:|:---:|:---:| | | | |

Live dashboard: boardclaude.com

Score Progression

BoardClaude audits itself and uses the results to improve. Here's the actual score progression across 18 iterations:

Each iteration runs 6 agents in parallel, identifies action items, implements fixes, validates with tsc + eslint, and re-audits. The dip at iteration 7 reflects a major architecture migration that temporarily reduced scores before recovering stronger.

Example Output

BoardClaude's first self-audit scored 68.4 / 100 (C+, MARGINAL) with 6 agents finding real issues:

• Boris (Architecture): "Strong architecture with real feedback loops, but zero tests and 20% format compliance"
• Cat (Product): "Genuinely novel feedback loop, but adoption path is too steep"
• Thariq (AI Innovation): "Smart multi-agent architecture, but emergent behavior is still theoretical"
• Lydia (Code Quality): "Clean TypeScript interfaces, but zero test coverage and missing performance optimizations"
• Ado (Docs/A11y): "Outstanding README, but not community-ready -- no CONTRIBUTING.md"
• Jason (Community): "Strong narrative, but hardcoded English strings limit global reach"

See the full audit JSON: examples/audit-example.json

Examples

The examples/ directory contains starter files and guides:

| File | Description | |------|-------------| | quickstart.md | Copy-pasteable install and first-audit walkthrough | | minimal-panel.yaml | Annotated minimal panel with 2 agents | | custom-agent.md | Step-by-step guide to creating a custom agent | | audit-example.json | Full audit output from a real self-evaluation |

Architecture

boardclaude/
├── .claude-plugin/plugin.json    # Plugin manifest
├── agents/                       # 7 agent persona files
├── commands/                     # 7 slash command definitions
├── skills/                       # 6 skill implementations
├── panels/                       # Panel YAML configurations
│   ├── hackathon-judges.yaml     # Default: 6-agent evaluation panel
│   └── templates/                # Additional panel templates
├── dashboard/                    # Next.js 15 web app
│   ├── src/app/                  # App Router pages
│   ├── src/components/           # React components
│   └── src/lib/types.ts          # Canonical TypeScript interfaces
└── .boardclaude/                 # Audit state (tracked in git)
    ├── state.json                # Project state
    ├── timeline.json             # Event timeline
    ├── action-items.json         # Action item ledger
    └── audits/                   # Audit result files

Panel Agents

The default hackathon-judges panel includes 6 specialized agents:

| Agent | Role | Weight | Model | |-------|------|--------|-------| | Boris | Architecture & Verification | 20% | Opus | | Cat | Product & User Impact | 18% | Opus | | Thariq | AI Innovation & Intelligence | 18% | Opus | | Lydia | Frontend/DX & Code Quality | 18% | Opus | | Ado | DevRel/Docs & Accessibility | 13% | Sonnet | | Jason | Community Impact & Integration | 13% | Sonnet |

How It Works

1. Audit -- Each agent independently evaluates the codebase against their weighted criteria
2. Synthesize -- A synthesis agent merges findings, identifies divergent opinions, prioritizes actions
3. Fix -- The fix command implements top-priority action items with validation gates
4. Re-audit -- A follow-up audit measures improvement and tracks score delta
5. Iterate -- Repeat until target score is reached or improvements plateau

Custom Panels

Create your own evaluation panel with a YAML config:

name: my-review-panel
type: professional
version: "1.0.0"
description: "Custom code review panel"

agents:
  - name: security-expert
    role: "Security & Vulnerability Analysis"
    weight: 0.40
    model: opus
    prompt_file: "agents/security.md"
    criteria:
      - name: vulnerability_scan
        weight: 0.50
      - name: auth_review
        weight: 0.50

  - name: perf-analyst
    role: "Performance & Optimization"
    weight: 0.60
    model: sonnet
    prompt_file: "agents/performance.md"
    criteria:
      - name: response_times
        weight: 0.50
      - name: resource_usage
        weight: 0.50

scoring:
  scale: 100
  passing_threshold: 70
  iteration_target: 85

Tech Stack

• Plugin: Claude Code Agent Teams, subagent orchestration
• Dashboard: Next.js 15, React 19, TypeScript, Tailwind CSS v4
• Visualization: Recharts, Framer Motion
• Deployment: Vercel

License

MIT