Claude Skills & Composable Personas

Reusable skills and composable system prompts for Claude Code.

Claude Launcher

Interactive system prompt and model selector for Claude Code.

Start Claude with your chosen persona and model in seconds:

# Interactive 2-step selection (persona → model)
$ cl

# Direct shortcuts (order-independent)
$ cl tdd opus        # Super TDD Developer + Opus
$ cl opt sonn        # Claude Code Optimizer + Sonnet
$ cl haik            # Generalist Robot + Haiku (default persona + model)

# Model-only (uses generalist-robot)
$ cl sonn
$ cl opus

Features:

• Order-independent shortcuts: cl tdd sonn = cl sonn tdd
• Frontmatter-based shortcuts (add your own personas instantly)
• Automatic skill loading via @ references
• Conflict detection with prominent warnings
• Exports CLAUDE_PERSONA for status line display

Setup:

alias cl='python3 /path/to/claude-skillz/claude-launcher/claude-launcher.py'

# Optional: Install fzf for interactive fuzzy search
brew install fzf  # macOS

See available personas below. You can add your own.

OpenCode Launcher

Generates OpenCode agents from the same personas used by Claude Launcher.

This repo contains system prompts that work with both Claude Code and OpenCode:

• Claude Launcher (cl) - for Claude Code
• OpenCode Launcher - generates agents for OpenCode

# Generate OpenCode agents (run once or after changing personas)
python3 opencode-launcher/generate-opencode-agents.py

# Then use OpenCode - agents appear in agent selector
opencode

See opencode-launcher/README.md for details.

Installation

Per-project

Add the plugin to your settings.json. See Claude Plugin Settings.

Globally Setup Marketplace

Local:

/plugin marketplace add file:///absolute/path/to/claude-skillz

GitHub:

/plugin marketplace add ntcoding/claude-skillz

Install Plugins

Interactive:

1. Run /plugin
2. Select Browse and install plugins
3. Select claude-skillz marketplace
4. Select desired plugin
5. Select Install now

Direct:

/plugin install <plugin-name>@claude-skillz

Available Skills

Skills are reusable behavioral instructions loaded into personas. Load them with @ references in your system prompts.

Research & Evidence

• independent-research - Research-driven investigation. Never guess—validate solutions before presenting. Use WebFetch, WebSearch, testing.
• confidence-honesty - Force honest confidence assessment. Express confidence as percentage, explain gaps, validate assumptions before presenting.

Communication & Output

• concise-output - Signal-over-noise. Eliminate verbose phrases, prioritize density. Every word must carry information.
• critical-peer-personality - Professional, skeptical communication. Challenge constructively, propose instead of asking, coach rather than serve.
• questions-are-not-instructions - Answer questions literally. Don't interpret as hidden instructions. STOP after answering, let user decide.

Code & Design

• software-design-principles - Object calisthenics, dependency inversion, fail-fast error handling, feature envy detection, intention-revealing naming.
• lightweight-implementation-analysis-protocol - Trace execution paths before implementing. Create lightweight diagrams. Prevent wasted effort from assumptions.
• lightweight-design-analysis - Systematic design review across 8 dimensions: Naming, Object Calisthenics, Coupling & Cohesion, Immutability, Domain Integrity, Type System, Simplicity, Performance.

Development Processes

• tdd-process - Strict TDD state machine: red-green-refactor with 11 enforced rules. Meaningful failures, minimum implementations, full verification.
• writing-tests - Principles for effective tests. Naming conventions, assertion best practices, comprehensive edge case checklists (based on BugMagnet).
• observability-first-debugging - Systematic debugging. Add instrumentation to gather specific data. Evidence before hypothesis.

Workflows & Tools

• switch-persona - Mid-conversation persona switching without restart. Lists personas, reads file, switches immediately.
• lightweight-task-workflow - Task list state machine for multi-session work. Tracks status, prevents auto-advancement, enforces state transitions.
• create-tasks - Convert requirements into actionable tasks following a structured template. Engineering-ready work items.

Specialized

• data-visualization - Build charts, graphs, dashboards. Visual execution, technical implementation, perceptual foundations, chart selection, layout algorithms, library guidance.
• typescript-backend-project-setup - NX monorepo setup for TypeScript backend projects optimized for AI-assisted development.

System Prompts

12 pre-built personas ready to use via cl shortcuts:

Composability

System prompts use @ references to load skills. See CLAUDE.md for composability guidelines and launcher docs for creating custom personas.a

Plugins

• track-and-improve - Capture mistakes and improvement opportunities with automatic 5 whys root cause analysis
• automatic-code-review - Automatic semantic code review on session stop with configurable project-specific rules. Auto-initializes with default rules, supports any language.

{
  "name": "claude-skillz",
  "owner": {
    "name": "Nick Tune",
    "email": "nick@ntcoding.co.uk"
  },
  "metadata": {
    "description": "Nick Tune's claude code skills",
    "version": "1.3.0",
    "homepage": "https://github.com/ntcoding/claude-skillz"
  },
  "plugins": [
    {
      "name": "task-check",
      "source": "./task-check",
      "description": "Verify task completion before finishing work. Catches incomplete work, bugs, and quality issues.",
      "version": "1.0.5",
      "category": "productivity",
      "keywords": [
        "task",
        "qa",
        "verification",
        "completion",
        "quality"
      ]
    },
    {
      "name": "automatic-code-review",
      "source": "./automatic-code-review",
      "description": "Automatic code review on session stop with configurable rules",
      "version": "1.0.4",
      "category": "development",
      "keywords": [
        "code-review",
        "quality",
        "hooks",
        "semantic-review"
      ]
    },
    {
      "name": "full-codebase-review",
      "source": "./full-codebase-review",
      "description": "Periodic full codebase review using automatic-code-review rules. Reviews entire codebase, not just modified files.",
      "version": "1.0.1",
      "category": "development",
      "keywords": [
        "code-review",
        "tech-debt",
        "codebase-analysis",
        "periodic-review",
        "quality"
      ]
    },
    {
      "name": "track-and-improve",
      "source": "./track-and-improve",
      "description": "Capture mistakes and improvement opportunities with automatic 5 whys analysis",
      "version": "1.0.0",
      "category": "productivity",
      "keywords": [
        "improvement",
        "tracking",
        "feedback",
        "learning"
      ]
    },
    {
      "name": "collaboration-modes",
      "source": "./collaboration-modes",
      "description": "Define collaboration style between Claude Code and user: step-by-step, deep-think, or pair programming",
      "version": "1.0.1",
      "category": "productivity",
      "keywords": [
        "collaboration",
        "workflow",
        "pair-programming",
        "modes"
      ]
    },
    {
      "name": "learn-from-prs",
      "source": "./learn-from-prs",
      "description": "Analyze PR feedback patterns from CodeRabbit, SonarQube, and reviewers to suggest config updates that catch issues locally",
      "version": "1.0.0",
      "category": "development",
      "keywords": [
        "pr-review",
        "feedback",
        "learning",
        "code-quality",
        "conventions"
      ]
    },
    {
      "name": "challenge-that",
      "source": "./challenge-that",
      "description": "Force critical evaluation of proposals from 5 adversarial perspectives (Skeptic, Pragmatist, Edge Case Hunter, Structural Critic, Root Cause Analyst)",
      "version": "1.0.0",
      "category": "productivity",
      "keywords": [
        "critical-thinking",
        "evaluation",
        "review",
        "proposals",
        "decisions"
      ]
    },
    {
      "name": "architect-refine-critique",
      "source": "./architect-refine-critique",
      "description": "Three-phase design review producing an ADR: Architect → Refiner → Critique → Interactive Review. Run /arc-review after to produce the ADR. Run /arc-prd to add architecture section to a PRD.",
      "version": "1.6.1",
      "category": "development",
      "keywords": [
        "architecture",
        "design-review",
        "refactoring",
        "ddd",
        "separation-of-concerns"
      ]
    },
    {
      "name": "session-optimizer",
      "source": "./session-optimizer",
      "description": "Analyze session transcripts for optimization opportunities using 4 parallel subagents",
      "version": "1.0.0",
      "category": "productivity",
      "keywords": [
        "session",
        "optimization",
        "analysis",
        "workflow"
      ]
    },
    {
      "name": "optimization-team",
      "source": "./optimization-team",
      "description": "Two-agent optimization team: researcher proposes, critic challenges, lead presents synthesized result",
      "version": "1.0.0",
      "category": "productivity",
      "keywords": [
        "optimization",
        "agent-teams",
        "research",
        "critique",
        "claude-code"
      ]
    },
    {
      "name": "claude-code-updates",
      "source": "./claude-code-updates",
      "description": "Detects Claude Code version updates and surfaces relevant new features. SessionStart hook nudges on updates, /whats-new command analyzes changelog and blog for applicable changes.",
      "version": "1.0.0",
      "category": "productivity",
      "keywords": [
        "updates",
        "changelog",
        "features",
        "version-tracking",
        "release-notes"
      ]
    },
    {
      "name": "development-skills",
      "source": "./",
      "description": "Random skills for various software development tasks",
      "version": "5.7.0",
      "category": "development",
      "keywords": [
        "tdd",
        "design",
        "workflow",
        "personality",
        "project-setup"
      ],
      "skills": [
        "./tdd-process",
        "./writing-tests",
        "./switch-persona",
        "./lightweight-implementation-analysis-protocol",
        "./lightweight-design-analysis",
        "./software-design-principles",
        "./critical-peer-personality",
        "./independent-research",
        "./concise-output",
        "./observability-first-debugging",
        "./data-visualization",
        "./confidence-honesty",
        "./questions-are-not-instructions",
        "./create-tasks",
        "./typescript-backend-project-setup",
        "./separation-of-concerns",
        "./tactical-ddd",
        "./fix-it-never-work-around-it"
      ]
    },
    {
      "name": "fetching-circleci-logs",
      "source": "./fetching-circleci-logs",
      "description": "Fetch CircleCI job logs via v1.1 API with step-level output and failure analysis",
      "version": "1.0.0",
      "category": "development",
      "keywords": [
        "circleci",
        "ci",
        "build-logs",
        "debugging",
        "pr-checks"
      ]
    }
  ]
}