v7.0.0 — Plugin-Only Distribution

Plugin-Only Distribution

The Claude Agent System is now distributed exclusively as a Claude Code plugin. The legacy script-based command system has been fully retired — 99 files removed, ~25,000 lines deleted.

Install

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

Breaking Changes

• Plugin renamed from pcc to cas — existing users must uninstall and reinstall
• Legacy system removed — .claude/commands/, .claude/workflows/, .claude/middleware/, .claude/agents/, and setup scripts deleted
• Script install path removed — setup-claude-agent-system.sh and setup-claude-agent-system.ps1 are gone

6 Skills — All Preserved and Unchanged

Skill	Purpose
/zk	Intelligent router — auto-picks the best skill for your task
/pcc	Parallel Claude Coordinator (Sonnet scouts + Opus implementers)
/pcc-opus	PCC with Opus scouts for maximum exploration quality
/hydra	Multi-task parallel swarm (Agent Teams, wave-based execution)
/review	7-agent code review swarm with opt-in auto-fix
/systemcc	Auto-routing workflow with Lyra AI optimization and triple review

Migration from Script Install

rm -rf .claude/commands .claude/workflows .claude/middleware .claude/agents
/plugin marketplace add Kasempiternal/Claude-Agent-System
/plugin install cas

Migration from Plugin pcc

/plugin uninstall pcc
/plugin install cas

See CHANGELOG.md for full details.

v5.1.0 - /zk Intelligent Router + Hydra Optimization

What's New

/zk — Intelligent Router Skill

Stop choosing between /pcc, /pcc-opus, and /hydra manually. Just use /zk and it picks for you.

/zk add a settings page                        # → PCC
/zk refactor the payment processing system      # → PCC-Opus
/zk fix auth; add dashboard; update API         # → Hydra

ZK uses a deterministic 4-step decision tree:

Step	Condition	Routes To
1	Multiple independent deliverables?	Hydra
2	Scale word + broad noun ("entire codebase")?	Hydra
3	High-stakes keyword + qualifying signal?	PCC-Opus
4	Everything else	PCC

When genuinely ambiguous, ZK asks you to choose instead of guessing.

Escape hatch: /pcc, /pcc-opus, and /hydra still work directly.

Hydra Context Optimization

Hydra orchestrator reduced from 951 to 371 lines (61% reduction). Delegates synthesis and wave prep to analyst teammates, keeping the orchestrator's context window lean for long multi-task sessions.

/review v2: Self-Contained Agents + Fix Phase

• All 6 review agents now defined as .md files in .claude/agents/ — no external plugin dependencies
• New opt-in fix phase: parallel fix agents resolve CRITICAL and MAJOR findings

README Reorganization

• Plugin skills shown first (/zk, /pcc, /pcc-opus, /hydra, /review)
• Script commands after (/plan-opus, /systemcc)
• Added full /hydra documentation to main README

---

Plugin Update

# If you already have the plugin, restart Claude Code to pick up the new /zk skill
# New installs:
/plugin marketplace add Kasempiternal/Claude-Agent-System
/plugin install pcc

Full Changelog: v5.0.0...v5.1.0

v6.1.0 - /review v2: Self-Contained Agents + Fix Phase

Highlights

The /review skill is now fully self-contained (no external plugin dependencies) and can automatically fix the issues it finds.

---

Self-Contained Review Agents

/review previously required feature-dev and pr-review-toolkit plugins to be installed. Now all 6 review agents are defined as .md files in .claude/agents/ and run via general-purpose subagents. The plugin works standalone.

Agent	File
Bug & Logic Reviewer	.claude/agents/review-bug-logic.md
Project Guidelines Reviewer	.claude/agents/review-guidelines.md
Silent Failure Hunter	.claude/agents/review-silent-failures.md
Comment Analyzer	.claude/agents/review-comments.md
Type Design Analyzer	.claude/agents/review-type-design.md
Test Coverage Analyzer	.claude/agents/review-test-coverage.md

Fix Phase (Replaces Simplification)

Phase 2 is no longer "simplification" — it's targeted fixes. After the review report, you choose:

• Fix CRITICAL and MAJOR — parallel fix agents resolve high-severity findings
• Fix ALL — also addresses MINOR findings
• Report only — keep the report, don't touch code

Fix agents are grouped by file (exclusive ownership, no merge conflicts) and make minimum changes per finding.

Improved Agent Prompts

All 6 agent prompts were enhanced over the original feature-dev / pr-review-toolkit versions:

Improvement	Detail
Swarm awareness	Each agent knows it's one of 6 with explicit "stay in your lane" boundaries
De-overlapped	bug-logic no longer checks conventions; guidelines no longer checks bugs
Standardized output	All agents produce consistent file:line + severity + suggested fix
Language patterns	Flags Go _ = err, Rust .unwrap(), Swift try?, Python except: pass
Priority filter	Comment agent filters out trivial findings
Type-design dual output	Per-type 1-10 ratings + per-finding file:line for fix agents

---

Install / Upgrade

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

Full Changelog: v6.0.0...v6.1.0

v6.0.0 - Plugin System, /review Skill & Decision Engine Overhaul

Plugin System & Code Review Swarm

Highlights

This release transforms Claude Agent System into a proper Claude Code plugin with 4 skills, adds a dedicated code review command, and overhauls the decision engine.

---

New: Plugin Architecture

The system is now installable as a Claude Code plugin at claude-agent-system-plugin/:

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

All skills (/systemcc, /pcc, /pcc-opus, /review) are now plugin skills with proper YAML frontmatter, argument hints, and model selection.

---

New: /review - Code Review Swarm

A review-only skill that deploys 6 parallel Anthropic review agents without modifying code:

/review                  # Review all uncommitted changes
/review staged           # Review only staged changes  
/review src/auth.ts      # Review specific files

Agent	Focus
Bug & Logic Reviewer	Security, crashes, logic errors
Project Guidelines Reviewer	Style, CLAUDE.md standards
Silent Failure Hunter	Swallowed exceptions, bad fallbacks
Comment Analyzer	Stale docs, misleading comments
Type Design Analyzer	Encapsulation, invariants
Test Coverage Analyzer	Test gaps, edge cases

Features:

• Health score (0-10) with deterministic severity-weighted formula
• Cross-agent correlation groups related findings
• Deduplicated consolidated report (orchestrator merges overlap)
• Agent verdicts table for quick scanning
• Opt-in simplification phase (2 parallel agents, only with user consent)

---

New: /pcc and /pcc-opus Skills

Parallel Claude Coordinator skills for implementation:

• /pcc - Sonnet scouts (fast, cost-efficient) + Opus implementers
• /pcc-opus - Opus scouts + Opus implementers (maximum quality)

Both feature:

• Dynamic agent count (2-6 based on complexity)
• Phase 3.5: Clarification & Questions before plan creation
• Editable plan files at .claude/plans/{task}.md
• Parallel implementation with focused file ownership
• Code simplification phase

---

Enhanced: Decision Engine

The /systemcc decision engine was overhauled:

• Two-phase workflow selection with confidence scoring
• 3-dimensional analysis (complexity, risk, scope) - simplified from 8-dimensional
• Accurate parallel execution documentation
• Fixed 5 critical issues around honesty and accuracy
• Cleaned up phantom commands and aligned documentation

---

Documentation

• Root README fully updated with /review, /pcc, /pcc-opus sections
• Plugin README rewritten with flow diagrams for all 4 skills
• Updated skill comparison tables and "When to Use" guide
• Plugin version bumped to 2.0.0

---

Installation

# macOS/Linux
curl -sSL https://raw.githubusercontent.com/Kasempiternal/Claude-Agent-System/main/setup-claude-agent-system.sh | bash

# Windows PowerShell
irm https://raw.githubusercontent.com/Kasempiternal/Claude-Agent-System/main/setup-claude-agent-system.ps1 | iex

Or install as a plugin:

/plugin marketplace add Kasempiternal/Claude-Agent-System

---

Upgrade

Re-run the installation script to get the latest version. Your cache and checkpoints are preserved.

Plan Opus & Agent Updates

Plan Opus Command & Enhanced Agents

🎯 Highlights

This release introduces the /plan-opus command for explicit planning control, adds missing agents for code review and simplification, and reorganizes documentation for clarity.

---

✨ New Features

/plan-opus Command

A new command for when you want full control over the planning process before execution:

• Configurable scout model - Sonnet by default (2-6 agents), configurable to Opus
• Editable plan file - Creates .claude/plans/{task}.md that you can review and modify
• Parallel implementation - 2-6 Opus agents work simultaneously
• Code simplification phase - Automatic cleanup after implementation

/plan-opus "refactor the authentication system"
# Creates plan → You review/edit → Confirm → Parallel execution

New Agents

Agent	Purpose
code-reviewer.md	Triple review with Senior Engineer, Lead Engineer, and Architect perspectives
code-simplifier.md	Post-implementation cleanup without changing functionality

---

📖 Documentation

• Reorganized README - /systemcc explained fully first, then /plan-opus
• Anti-vibe coding philosophy - Clarified that both commands enforce structure (triple review, validation, learning)
• Accurate native plan mode comparison - Fixed inaccuracies about Claude Code's native /plan

---

📦 Installation

# macOS/Linux
curl -sSL https://raw.githubusercontent.com/Kasempiternal/Claude-Agent-System/main/setup-claude-agent-system.sh | bash

# Windows PowerShell
irm https://raw.githubusercontent.com/Kasempiternal/Claude-Agent-System/main/setup-claude-agent-system.ps1 | iex

Add --global or -Global to install system-wide.

---

🔄 Upgrade

Re-run the installation script to get the latest version. Your cache and checkpoints are preserved.

v4.0.0 - Persistent Caching & Major Cleanup

Highlights

Persistent Analysis Cache

• Project analysis cached in ~/.claude/cache/ for instant startup
• Smart invalidation on git commits or file changes
• Cross-session persistence without polluting target repos

Architecture Cleanup

• Consolidated to .claude/ as single source of truth
• Removed duplicate root directories
• Deleted legacy ClaudeFiles system
• Simplified setup scripts (~450 lines removed)

New Cache Commands

• --reanalyze: Force fresh analysis
• --clear-cache: Clear cache for current repo

What's New

• Instant startup on subsequent sessions (cached analysis)
• Per-repo isolation in ~/.claude/cache/{repo-hash}/
• Auto-refresh when project changes significantly
• Checkpoint system for session resumption
• 32,000+ lines of dead code removed

Breaking Changes

None - fully backward compatible.

Installation

# macOS/Linux
curl -sSL https://raw.githubusercontent.com/Kasempiternal/Claude-Agent-System/main/setup-claude-agent-system.sh | bash

# Windows PowerShell
irm https://raw.githubusercontent.com/Kasempiternal/Claude-Agent-System/main/setup-claude-agent-system.ps1 | iex

v2.1.0 - Build Configuration Auto-Detection

Build Configuration Auto-Detection

Automatically detects and applies project build rules (Makefile, CI/CD, linters) to ensure all generated code passes your pipeline on
first commit.

Key Features

• Auto-scans Makefile, .gitlab-ci.yml, .github/workflows, and config files
• Extracts formatting rules (black, prettier, isort with exact flags)
• Applies linting standards (flake8, eslint, mypy with your ignores)
• Supports Python, JavaScript, Go, and more
• Zero configuration - just works!

Example

If your Makefile has black --line-length 100 . and flake8 --ignore E501,E203, all generated Python code automatically follows those
exact rules.

Benefits

✅ No more pipeline failures
✅ Code passes CI/CD checks on first commit
✅ Team consistency automatically enforced
✅ Saves time on manual formatting

New Files: build-config-detector.md, 03-BUILD-CONFIG.md, Makefile.example
Version: 2.0.0 → 2.1.0 (minor release, backward compatible)

Or even shorter:

Auto-detects build configuration (Makefile, CI/CD, linters) and ensures all generated code follows your project's formatting and linting
rules. No more pipeline failures - code passes checks on first commit. Supports Python, JavaScript, Go. Zero configuration needed.

v2.0.0 - Enhanced Review & Memory System

[2.0.0] - 2025-01-10

🎉 Major Release - Enhanced Review & Memory System

This is a significant update that adds enterprise-grade code review and intelligent memory management to the Claude Agent System.

Added

• Triple Code Review System - Three specialized reviewers run in parallel after implementation:
  • Senior Engineer reviewer (code quality, best practices, clean code)
  • Lead Engineer reviewer (architecture, technical debt, scalability)
  • Architect reviewer (system integration, enterprise patterns)
  • All reviews complete in 5 minutes max with auto-fix for critical issues
• Automatic Memory Updates - Every session now automatically updates all memory files:
  • New patterns discovered → CLAUDE-patterns.md
  • Architecture decisions → CLAUDE-decisions.md
  • Issues resolved → CLAUDE-troubleshooting.md
  • User corrections → CLAUDE-dont_dos.md (NEW)
• User Preference Learning - New CLAUDE-dont_dos.md file:
  • Captures what you DON'T want when you say "no", "stop", or "don't"
  • Prevents repeating unwanted behaviors
  • Permanent preference memory across sessions
• Installation Verification - New test-enhanced-system.sh script:
  • Verifies all components are correctly installed
  • Checks for missing files and provides clear error messages
  • Quick validation after setup
• Enhanced Workflow Phases:
  • Phase 8: Post-Execution Triple Review
  • Phase 9: Memory Bank Auto-Update
  • Phase 10: Brief Summary (no fluff, no unnecessary docs)

Enhanced

• 8-Dimensional Analysis Engine (upgraded from 5-dimensional):
  • Added: Code Minimalism Score (modify vs create analysis)
  • Added: Security Sensitivity (auth, database, encoding detection)
  • Added: Pattern Reusability (existing patterns to leverage)
  • Existing: Technical Complexity, Scope Impact, Risk Assessment, Context Load, Time Pressure
• Systemcc Modules - Added new documentation modules:
  • 10-POST-REVIEW.md - Review system documentation
  • 11-MEMORY-UPDATE.md - Memory auto-update documentation
• Setup Script - Enhanced installation:
  • Now installs code review agents
  • Creates all new memory files including dont_dos.md
  • Verifies post-execution review workflow
  • Better error handling and path detection

Changed

• Workflow Enforcement - Updated to include review + memory phases
• Memory Bank Structure - Now includes 5 files (added dont_dos.md)
• Project Structure - New directories for agents and post-execution-review workflow

Fixed

• README Accuracy - Corrected all references from 5 to 8 dimensions
• Documentation Consistency - All docs now match actual implementation
• Flowchart Updates - Decision engine flowchart shows all 8 dimensions

Technical Details

• 24 files changed, 3,539 insertions(+), 293 deletions(-)
• 9 new files created (agents, workflows, memory)
• Backward compatible - all existing workflows still function

---

[1.0.0] - 2025-01-18

Initial Release

The first public release of the Claude Agent System.

Core Features

• 5-Dimensional Task Analysis - Technical Complexity, Scope Impact, Risk Assessment, Context Load, Time Pressure
• Multiple Workflow Support:
  • Complete System (6-agent validation)
  • Orchestrated (3-agent streamlined)
  • Phase-based (large codebase handling)
  • Agent OS (project initialization)
  • AI Dev Tasks (PRD-based development)
  • Anti-YOLO Web (wireframe-first web development)
• Memory Bank System - Persistent learning across sessions:
  • Active context tracking
  • Pattern recognition
  • Decision logging
  • Troubleshooting database
• Lyra AI Optimization - 4-D prompt enhancement methodology
• Security Integration - PromptSecure-Ultra scanner
• Intelligent Workflow Selection - Automatic routing based on task analysis
• CCPM Integration - Project management with GitHub/GitLab
• Batch Optimization - Parallel operation detection and execution

Components

• Modular systemcc command (9 modules)
• 6 specialized workflows
• Middleware for decision engines and optimization
• File organization system (.claude/files/)
• Memory persistence system
• Git workflow integration

---