Codingbuddy

Prove your AI coding is actually improving.

Codingbuddy is a multi-AI MCP server that orchestrates 37 specialist agents across 9 AI tools — and measures the impact on your code quality with every session.

What's New in v5.6.0 — HUD Statusbar Wave

• Breathing Buddy Face — the buddy face reacts to your session phase (idle / thinking / active / blocked / victory) so the status bar feels alive.

• Cost Velocity Indicator — session spend-rate badge with 🔥 / ↗ / → / 💤 trend glyphs tells you whether you're on a slow planning pass or a hot refactor burn.

• Cache Savings Badge — quantifies prompt-cache discounts as 💰$N.NN saved so you see the value of caching in real time.

• Mode Rainbow Coloring — per-mode ANSI truecolor gradients (PLAN ◇ / ACT ◆ / EVAL ◈ / AUTO ◊) with NO_COLOR env support for CI and greyscale terminals.

• Smart Context Bar — visual [████░░░░░░] 42% progress bar replaces plain Ctx:42% text with warning and danger thresholds.

• Adaptive Layout Engine — HUD fits narrow terminals gracefully with fit_segments adaptive truncation.

• Rate-limit Severity Icons — visual warnings when you approach API rate limits.

Related MCP server: Multi-MCP

Session Impact Report

At the end of every session, codingbuddy shows you what actually happened:

┌─────────────────────────────────────────────────┐
│            Session Impact Report                │
├─────────────────────────────────────────────────┤
│                                                 │
│  Issues prevented           12                  │
│    Security                  4  (2 critical)    │
│    Accessibility             3                  │
│    Performance               2                  │
│    Code Quality              3                  │
│                                                 │
│  Agents dispatched           8                  │
│  Checklists generated        5                  │
│  Mode transitions            PLAN → ACT → EVAL │
│  Context decisions          14                  │
│                                                 │
└─────────────────────────────────────────────────┘

Most AI rules tools tell you to "code better." Codingbuddy proves you did.

Quick Start

# 1. Install
npm install -g codingbuddy

# 2. Initialize your project
npx codingbuddy init

# 3. Add to your AI tool's MCP config

{
  "mcpServers": {
    "codingbuddy": {
      "command": "npx",
      "args": ["codingbuddy", "mcp"]
    }
  }
}

Start coding with PLAN, ACT, EVAL, or AUTO keywords.

Claude Code users: Install as a plugin for the full experience:

# 1. Add marketplace
claude marketplace add JeremyDev87/codingbuddy

# 2. Install plugin
claude plugin install codingbuddy@jeremydev87

The plugin adds session-aware hooks that power Impact Reports, buddy greetings, onboarding tours, achievement badges, and adaptive performance mode. Update to the latest version with claude plugin install codingbuddy@jeremydev87.

See the Plugin Guide for details.

Full Getting Started Guide →

Why Codingbuddy?

Measurable results, not promises. Session Impact Reports show issues prevented, agents dispatched, and quality improvements — backed by data from every tool call.

37 specialists, not one generalist. Security engineers, accessibility experts, performance specialists, architects — each with domain expertise that a single AI cannot match.

One ruleset, 9 AI tools. Same quality standards whether you use Cursor, Claude Code, GitHub Copilot, or any other supported tool. Switch tools without losing consistency.

Quality gates that ship. The PLAN → ACT → EVAL cycle iterates until Critical=0 and High=0. Production-ready code, not "good enough" code.

What's New in v5.4.0

Question-First Planning — Codingbuddy now asks before it plans. Ambiguous prompts trigger a clarifying question, and clear prompts walk through Discover→Design→Plan stages with explicit user confirmation at each step.

Council Scene — PLAN, EVAL, and AUTO modes open with a visible council scene showing which specialist agents are assembled and ready to collaborate.

Permission Forecast — Before execution, codingbuddy surfaces what permission classes (repo-write, network, external) will be needed, so you can prepare approval bundles in advance.

Execution Gate — When the planning stage hasn't advanced past discovery, specialist dispatch is suppressed to avoid premature work.

Council State Pipeline — Real-time council badges show agent handoffs, stage transitions, and blockers as they happen during tool execution.

Supported AI Tools

All Setup Guides →

How It Works

Codingbuddy enforces a quality-driven development cycle:

    PLAN                    ACT                     EVAL
 ┌──────────┐          ┌──────────┐          ┌──────────────┐
 │ Architect │          │Developer │          │  Code Review  │
 │  designs  │───────▶  │implements│───────▶  │  + Parallel   │
 │  approach │          │ with TDD │          │  Specialists  │
 └──────────┘          └──────────┘          └──────┬───────┘
                                                    │
                                          ┌─────────┴─────────┐
                                          │                   │
                                    Critical > 0?       Critical = 0
                                    High > 0?           High = 0
                                          │                   │
                                          ▼                   ▼
                                    Back to PLAN        Ship with
                                                       confidence

AUTO mode runs the full cycle autonomously:

AUTO: Implement JWT authentication with refresh tokens
→ Plans architecture with security requirements
→ Implements with TDD
→ Reviews: security, performance, accessibility, code quality
→ Iterates until production-ready

Impact Telemetry

Codingbuddy tracks what its agents do during your session and generates a summary at the end.

What It Tracks

How It Works

1. Every MCP tool call logs impact events to the session

2. Events are categorized by domain (security, accessibility, performance, quality) and severity

3. At session end, get_session_impact generates the summary report

No configuration needed. Impact telemetry is always on.

Agents

37 specialist agents organized in a 3-tier system.

Mode Agents (4)

Orchestrate the development workflow.

Primary Agents (18)

Core implementation and review roles.

Specialist Agents (13)

Domain experts dispatched in parallel during EVAL.

Utility Agents (2)

Full Agent Reference →

Built-in Skills

Reusable workflows that enforce consistent development practices.

Full Skills Library →

Configuration

Create codingbuddy.config.json in your project root:

{
  "language": "en",
  "verbosity": "compact",
  "ai": {
    "defaultModel": "claude-sonnet-4-20250514"
  }
}

Configuration Reference →

Terminal Dashboard (TUI)

Real-time visualization of agent activity, task progress, and workflow state.

npx codingbuddy tui

Multi-session support, responsive layout, and zero overhead on MCP response times.

TUI Guide →

Documentation

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

License

MIT © Codingbuddy