Skip to content

Kasempiternal/Claude-Agent-System

Repository files navigation

Claude Agent System

Turn Claude into your personal development team. Two powerful commands that handle everything - from deep planning through implementation to deployment, with automatic code review and continuous learning.

Quick Start

Choose your installation method:

Method Best For Commands You Get
Plugin Quick install, easy updates /systemcc, /pcc, /pcc-opus
Script Full system with all modules /systemcc, /plan-opus, + workflows

Option 1: Plugin Install (Recommended)

/plugin marketplace add Kasempiternal/Claude-Agent-System
/plugin install pcc

Done! You now have /systemcc, /pcc, and /pcc-opus.

Option 2: Script Install (Full System)

macOS/Linux:

# Install for current project
curl -sSL https://raw.githubusercontent.com/Kasempiternal/Claude-Agent-System/main/setup-claude-agent-system.sh | bash

# Install globally (available in ALL projects)
curl -sSL https://raw.githubusercontent.com/Kasempiternal/Claude-Agent-System/main/setup-claude-agent-system.sh | bash -s -- --global

Windows (PowerShell):

# Install for current project
irm https://raw.githubusercontent.com/Kasempiternal/Claude-Agent-System/main/setup-claude-agent-system.ps1 | iex

# Install globally
.\setup-claude-agent-system.ps1 -Global

This installs the full system with all 13 systemcc modules, workflows, and middleware.


Anti-Vibe Coding Philosophy

"Vibe coding" = typing a prompt, accepting whatever the AI outputs, hoping it works.

This system is the opposite. Both commands enforce structure:

What Vibe Coding Does What This System Does
Blindly accepts AI output Triple code review (3 parallel reviewers)
No validation Build config detection + linting enforcement
No learning Session memory - learns your patterns and mistakes
No quality gates Decision engine with complexity/risk/scope analysis
Hope it works Post-execution validation + auto-fix critical issues

The /systemcc Command

The main command. Handles everything automatically with built-in quality gates.

/systemcc "what you want to do"

How It Works

When you run /systemcc, the system:

  1. Analyzes your project - Scans structure, tech stack, and conventions (cached for instant startup)
  2. Optimizes your request - AI enhancement for clarity and completeness
  3. Detects build configuration - Auto-scans Makefile/CI/CD for code standards
  4. Selects the best workflow - Picks between 3-agent, 6-agent, or specialized flows
  5. Executes automatically - All phases run without manual intervention
  6. Reviews the code - 3 parallel reviewers check quality
  7. Shows a brief summary - What changed and why

Examples

# Simple fixes - Fast 3-agent workflow
/systemcc "fix the login button color"

# Complex features - Full 6-agent system
/systemcc "add user authentication with JWT"

# Web projects - Automatic wireframing first
/systemcc "create contact form page"
# Shows ASCII wireframe -> You approve -> Builds HTML/CSS/JS

# Batch operations - Auto-detected
/systemcc "create CRUD for users, posts, comments"
# Groups operations -> Reduced tool switching

Command Options

/systemcc "your task"              # Standard execution
/systemcc --debug "your task"      # Show AI decision-making process
/systemcc --secure "task"          # Enhanced security scanning
/systemcc --reanalyze "task"       # Force fresh analysis (ignore cache)
/systemcc --clear-cache            # Clear cache for current repo

Key Features

Intelligent Workflow Selection

The system analyzes your request across three dimensions:

  • Complexity - Simple fix or complex architecture change?
  • Risk - Low-risk styling or high-risk security changes?
  • Scope - Single file, multiple files, or system-wide?

Then automatically picks the right workflow. No manual selection needed.

Build Configuration Auto-Detection

The system automatically detects and applies your project's build rules:

  • Scans Makefile, CI/CD files, linting configs
  • Extracts formatting rules (black, prettier, isort)
  • Applies linting standards (flake8, eslint, mypy)
  • Ensures all generated code passes your pipeline on first commit

If your Makefile has black --line-length 100, all Python code automatically uses 100-character lines.

Triple Code Review

After implementation, three specialized reviewers run in parallel:

  • Senior Engineer - Checks code quality, best practices, clean code
  • Lead Engineer - Reviews architecture, technical debt, scalability
  • Architect - Validates system integration, enterprise patterns

All three run simultaneously (5 minutes max). Critical issues are auto-fixed immediately.

Persistent Analysis Cache

Project analysis is cached in ~/.claude/cache/ for instant startup across sessions:

  • First run - Full analysis, cached to disk
  • Subsequent runs - Loads cache in milliseconds
  • Auto-refresh - Cache invalidates on git commits or major file changes
  • Zero pollution - No files created in your repository
First run:  🔍 Analyzing... (5-10 seconds) → 💾 Cached
Next runs:  ✅ Loaded cache (instant)
After git commit: 🔄 Refreshing cache...

Session-Based Learning

Within each session, the system learns:

  • Your patterns - Coding style, naming conventions, preferences
  • Your decisions - Architecture choices, technology selections
  • Your corrections - What you DON'T want (captured when you say "no" or "stop")

Progressive Context Management

The system intelligently manages context to handle larger codebases:

  • MINIMAL loading (60% reduction) - Simple tasks load only headers & signatures
  • STANDARD loading (30% reduction) - Medium tasks load summaries & key patterns
  • FULL loading - Complex tasks load complete documentation
  • Auto-checkpoints - Never lose progress, resume from interruptions

Anti-YOLO Web Development

For web projects, the system creates an ASCII wireframe first:

/systemcc "create a contact form"

Creating ASCII Wireframe:
┌─ Contact Us ─────────────────────────┐
│ Get in touch with our team           │
├─────────────────────────────────────┤
│ Name:     [________________]         │
│ Email:    [________________]         │
│ Subject:  [▼ General Inquiry]        │
│ Message:  [________________]         │
│           [________________]         │
│ ──────────────────────────────────── │
│ [Submit Message] [Clear Form]        │
└─────────────────────────────────────┘

Does this layout look right?
Type 'yes' to build HTML/CSS, or request changes.

Why this works:

  • 90% fewer revisions - Fix layout in wireframe stage (cheap) not code stage (expensive)
  • Token efficient - ASCII uses 10x fewer tokens than HTML mockups
  • No surprises - See exactly what you'll get before any code is written

Available Workflows

The system automatically chooses from these workflows:

Orchestrated (3-Agent System)

  • Orchestrator - Plans and coordinates
  • Developer - Implements the solution
  • Reviewer - Quality checks and testing

Best for: Bug fixes, simple features, refactoring

Complete System (6-Agent Validation)

  • Planner - Strategic analysis and architecture
  • Executer - Implementation and coding
  • Verifier - Logic and integration testing
  • Tester - Quality assurance and edge cases
  • Documenter - Code documentation and guides
  • Updater - Version control and deployment

Best for: New features, complex changes, critical systems

Phase-Based (Large Codebase Handler)

  • Breaks massive tasks into focused phases
  • Maintains context quality across large projects
  • Checkpoint system prevents context loss

Best for: Enterprise codebases, major refactors, system migrations

Agent OS (Project Setup)

  • Analyzer - Assesses current project state
  • Architect - Designs standards and structure
  • Builder - Implements foundation
  • Documenter - Creates project documentation

Best for: New project setup, standards implementation

AI Dev Tasks (PRD-Based Development)

  • PRD Creation - Requirements and specifications
  • Task Generation - Detailed work breakdown
  • Implementation - Feature building with validation

Best for: Product features, user stories, MVP development

Anti-YOLO Web

  • ASCII wireframe creation → approval → HTML implementation

Best for: UI components, forms, dashboards, landing pages

The Decision Engine

Here's what happens behind the scenes when you run /systemcc:

User: /systemcc "your request"
         │
         ▼
┌─────────────────────────────────────┐
│  PROJECT ANALYSIS                   │
│  Analyze structure, detect patterns │
└────────────┬────────────────────────┘
             │
             ▼
┌─────────────────────────────────────┐
│  SECURITY PRE-SCAN (if needed)      │
│  Check for injection, block threats │
└────────────┬────────────────────────┘
             │
             ▼
┌─────────────────────────────────────┐
│  LYRA AI PROMPT OPTIMIZATION        │
│  Deconstruct → Diagnose → Develop   │
└────────────┬────────────────────────┘
             │
             ▼
┌─────────────────────────────────────┐
│  BUILD CONFIG DETECTION             │
│  Makefile, CI/CD, linters           │
└────────────┬────────────────────────┘
             │
             ▼
┌─────────────────────────────────────┐
│  3-DIMENSIONAL ANALYSIS             │
│  Complexity × Risk × Scope          │
└────────────┬────────────────────────┘
             │
             ▼
┌─────────────────────────────────────┐
│  WORKFLOW SELECTION                 │
│  Pick best workflow for the task    │
└────────────┬────────────────────────┘
             │
             ▼
        EXECUTION
             │
   ┌─────────┼─────────┐
   ▼         ▼         ▼
┌─────┐ ┌─────┐ ┌─────────┐
│BATCH│ │ANTI │ │STANDARD │
│MODE │ │YOLO │ │WORKFLOWS│
└─────┘ └─────┘ └─────────┘

Debug Mode

Want to see how the system makes decisions?

/systemcc --debug "add user authentication"

ANALYSIS RESULTS:
├─ Complexity: complex (auth, security keywords)
├─ Risk: high (authentication detected)
└─ Scope: multi (auth, middleware, database)

DECISION: Complete 6-Agent System
   Confidence: 85% (High complexity + high risk)
   Security scan: enabled

Executing Complete System workflow...

The /plan-opus Command

For when you want explicit control over the planning process.

/plan-opus "task description"

Why This Command Exists

Claude Code has a native "plan mode" (/plan), but the community discovered a limitation: it uses Haiku as the code scout. While Haiku is efficient and fast, it's also the least capable model in the Claude family. For complex codebases, you may want smarter models doing the exploration.

/plan-opus was created to give you more control over the planning process with configurable models:

Aspect Native Plan Mode /plan-opus
Scout Model Haiku (2-3 agents) Sonnet by default (2-6 agents, configurable to Opus)
Plan Visibility Shown to user Written to editable .md file
User Approval Yes, before execution Yes, with ability to edit the plan first
Parallelization Limited (Claude Code rarely parallelizes) Aggressive (multiple agents per phase)
Implementation Sequential 2-6 parallel Opus agents
Post-Cleanup None 2-6 code simplifier agents

Configurable Scout Model

By default, scouts use Sonnet to balance intelligence and token cost. But if you want maximum exploration quality, you can switch scouts to Opus.

Edit .claude/commands/plan-opus.md, line 30:

# Default (token-efficient):
...using the Task tool with `subagent_type='Explore'` and `model='sonnet'`

# Maximum quality (change 'sonnet' to 'opus'):
...using the Task tool with `subagent_type='Explore'` and `model='opus'`

Why Sonnet is the default: Running 2-6 Opus scouts + 2-6 Opus implementers + 2-6 Opus simplifiers can consume significant tokens. Sonnet scouts are smart enough for exploration while keeping costs reasonable. Switch to Opus scouts only for particularly complex codebases.

How It Works

/plan-opus follows an orchestrator pattern - Opus coordinates everything but delegates actual work to specialized agents:

/plan-opus "your complex task"
         │
         ▼
┌─────────────────────────────────────┐
│  Phase 1: TASK UNDERSTANDING        │
│  Clarify requirements if needed     │
└────────────┬────────────────────────┘
             │
             ▼
┌─────────────────────────────────────┐
│  Phase 2: PARALLEL EXPLORATION      │
│  2-6 Sonnet scouts explore codebase │
│  (Architecture, Features, Tests...) │
└────────────┬────────────────────────┘
             │
             ▼
┌─────────────────────────────────────┐
│  Phase 3: SYNTHESIS                 │
│  Combine findings, verify key files │
└────────────┬────────────────────────┘
             │
             ▼
┌─────────────────────────────────────┐
│  Phase 4: PLAN CREATION             │
│  Write detailed plan to file        │
│  .claude/plans/{task-slug}.md       │
└────────────┬────────────────────────┘
             │
             ▼
┌─────────────────────────────────────┐
│  Phase 5: YOUR REVIEW               │
│  ⏸️  STOPS HERE - You edit the plan │
│  Confirm when ready to proceed      │
└────────────┬────────────────────────┘
             │
             ▼
┌─────────────────────────────────────┐
│  Phase 6: PARALLEL IMPLEMENTATION   │
│  2-6 Opus agents work in parallel   │
└────────────┬────────────────────────┘
             │
             ▼
┌─────────────────────────────────────┐
│  Phase 7: VERIFICATION              │
│  Tests + Code review                │
└────────────┬────────────────────────┘
             │
             ▼
┌─────────────────────────────────────┐
│  Phase 8: SIMPLIFICATION            │
│  2-6 agents clean up the code       │
└─────────────────────────────────────┘

The Plan File

Unlike automatic workflows, /plan-opus creates an actual file you can review and edit:

# Implementation Plan: Add User Authentication

Created: 2024-01-15
Status: PENDING APPROVAL

## Summary
Add JWT-based authentication with login, logout, and session management.

## Parallelization Strategy

| Stream | Focus | Files | Can Parallel With |
|--------|-------|-------|-------------------|
| Stream A | Database | migrations/, models/ | B, C |
| Stream B | API Routes | routes/auth.ts | A, C |
| Stream C | Middleware | middleware/auth.ts | A, B |

## Implementation Phases
...

---
**USER: Please review this plan. Edit any section directly, then confirm to proceed.**

You can:

  • Edit the plan directly in your editor
  • Add or remove phases
  • Change file assignments
  • Adjust the parallelization strategy
  • Then confirm to execute

Example Usage

# Complex feature - want to review the plan
/plan-opus "add real-time notifications with WebSocket"

# System creates plan at .claude/plans/add-real-time-notifications.md
# You review, maybe edit the WebSocket library choice
# You confirm
# 4 Opus agents implement in parallel
# Done!

When to Use Each Command

Situation Use This
Quick fixes, bug fixes /systemcc
Simple features /systemcc
Most everyday tasks /systemcc
Complex refactors /plan-opus
Architecture changes /plan-opus
When you want to see/edit the plan first /plan-opus

Project Structure

When installed, the system adds this structure:

your-project/
└── .claude/
    ├── commands/              # Command definitions
    │   └── systemcc/          # Modular systemcc modules
    ├── agents/                # Code reviewers and simplifiers
    ├── workflows/             # Workflow implementations
    │   ├── anti-yolo-web/
    │   ├── complete-system/
    │   ├── orchestrated-only/
    │   ├── phase-based-workflow/
    │   ├── agent-os/
    │   └── ai-dev-tasks/
    └── middleware/            # AI optimization systems

Data is stored separately in your home directory (never in your project):

~/.claude/
├── cache/                     # Persistent analysis cache (per-repo)
├── checkpoints/               # Session resumption data
└── temp/                      # Workflow temp files (auto-deleted)

Installation Options

Feature Local (default) Global (--global)
Available in Current project only All projects
Install location ./.claude/ ~/.claude/
Use case Project-specific setup Always-available commands

Tip: Use --global if you want /systemcc available everywhere.


Other Commands

/help                              # Show all available commands
/analyzecc                         # Manual project analysis

Contributing

Contributions are welcome! See CONTRIBUTING.md for guidelines.


Community

Built from real-world experiences shared by developers:


License

MIT License - see LICENSE for details.

About

No description, website, or topics provided.

Resources

License

Contributing

Stars

Watchers

Forks

Packages

No packages published

Contributors 5