GoodVibes Plugin
Plug in. Receive good vibes.
A Claude Code plugin that replaces native tools with token-efficient precision equivalents, adds 73 MCP tools across 6 engines, and orchestrates 11 specialized agents with persistent cross-session memory.
Installation
Prerequisites: Node.js (v18+) and tmux must be installed.
Supported platforms: Linux (x64, ARM64), macOS (x64, Apple Silicon), Windows (x64, ARM64, ia32). Native AST binaries are included for all platforms.
claude plugin marketplace add mgd34msu/goodvibes-plugin
claude plugin install goodvibes@goodvibes-market
After installation, run the Setup hook to pre-write CLAUDE.md chain files:
claude --init-only
This ensures all GoodVibes instruction files are in place before your first session. On each session start, the SessionStart hook:
- Detects your project stack (frameworks, languages, tools)
- Analyzes git status (branch, uncommitted changes)
- Checks project health (missing dependencies, build issues)
- Verifies CLAUDE.md chain files (writes/appends any that are missing)
- Injects project context into Claude's system message
Set your output style:
/output-style goodvibes:vibecoding # Interactive mode
/output-style goodvibes:justvibes # Autonomous mode
At a Glance
| Component | Count | What You Get |
|---|
| Agents | 11 | Specialized roles (Opus/Sonnet) for engineering, review, testing, architecture, deployment, integration, planning |
| Skills | 25 | Tiered knowledge modules: protocol, orchestration, outcome, quality |
| MCP Tools | 73 | Token-efficient tools across 6 specialized engines |
| Hooks | 11 | Lifecycle automation (tool redirection, context injection, error recovery, setup) |
| Output Styles | 2 | Interactive (vibecoding) or fully autonomous (justvibes) |
| Templates | 3 | Production scaffolds |
Why GoodVibes?
Token Efficiency
Token consumption in AI coding sessions follows a layered pattern: individual operations add tokens, round trips resend conversation context, sessions accumulate state, and knowledge either persists or gets rediscovered. GoodVibes optimizes all seven layers.
Layer 1: Per-Operation Savings
Native tools return maximum output regardless of need. Precision tools let you request exactly the detail level required.
Note: Token estimates below are for typical small-to-medium files (~50-100 lines). Savings scale linearly with file size (e.g., a 500-line file would be ~5,000 tokens native vs. the same low precision overhead).
| Operation | Native Tool | Precision Tool | Savings |
|---|
| Check if a file exists | Read returns full content (~500+ tokens) | precision_read with count_only (~15 tokens) | ~97% |
| Count files matching a pattern | Glob returns all paths (~200+ tokens) | precision_glob with count_only (~15 tokens) | ~92% |
| Check if a pattern exists in code | Grep returns all matches with context (~300+ tokens) | precision_grep with count_only (~15 tokens) | ~95% |
| Re-read an unchanged file | Read returns full content again (~500+ tokens) | precision_read returns cache hit (~20 tokens) | ~96% |
| Get function signatures from a file | Read returns entire file (~500+ tokens) | precision_read with symbols extract (~50 tokens) | ~90% |
Mechanisms:
- Verbosity levels (4-6 per tool):
count_only, files_only, minimal, standard, verbose. Tools default to minimal output automatically — precision_edit defaults to minimal, precision_grep to files_only, precision_glob to paths_only. Savings are automatic even without explicit requests.
- Extract modes (
precision_read): content, outline, symbols, ast, lines. Get function signatures (~50 tokens) instead of full file content (~500+ tokens). 75-95% savings.
- Token budget pagination: Large results auto-paginate to stay within a specified token limit. Prevents single responses from consuming disproportionate context.
- AST pattern matching (
precision_edit): More precise than regex, fewer false positives, fewer failed edits requiring retry.
Layer 2: Per-Round-Trip Savings
Every API call resends the entire conversation (system prompt + tool definitions + all messages). Fewer calls = less overhead.
mgd34msu