ghengis-skills

   ██████╗ ██╗  ██╗███████╗███╗   ██╗ ██████╗ ██╗███████╗
  ██╔════╝ ██║  ██║██╔════╝████╗  ██║██╔════╝ ██║██╔════╝
  ██║  ███╗███████║█████╗  ██╔██╗ ██║██║  ███╗██║███████╗
  ██║   ██║██╔══██║██╔══╝  ██║╚██╗██║██║   ██║██║╚════██║
  ╚██████╔╝██║  ██║███████╗██║ ╚████║╚██████╔╝██║███████║
   ╚═════╝ ╚═╝  ╚═╝╚══════╝╚═╝  ╚═══╝ ╚═════╝ ╚═╝╚══════╝
                     S K I L L S

Skills are lightweight — they load on-demand and don't bloat your context window. Unlike MCP servers that inject tool schemas into every message, skills activate only when relevant, adding zero overhead the rest of the time.

What's New (v1.7.0)

• Evolving Cognition (evolving-cognition) — ASI-Evolve-inspired pattern for agents that learn from measurable outcomes. Covers fitness signal design, cognition store schema, UCB1 retrieval, Analyzer prompts, poison mitigations (contradiction flags, confidence decay, retirement thresholds), and audit loops. Auto-fires on "feedback loop", "agents getting smarter", "learn from outcomes".
• Paper to Code (paper-to-code) — Turn any research paper, engineering blog, or technical doc into shipping code. Strategic reading order, contribution mapping tables, explicit "what NOT to adopt" phase, spec-delta writing. Auto-fires on arxiv URLs, paper links, "apply this research".
• Multi-Stage Verification (enhanced completion-enforcer) — New Check 6 adds the ASI-Evolve 3-tier evaluation pattern: proxy (lint/type-check) -> functional (tests) -> full (manual/UI verification). Catches "tests pass but I didn't actually try it" false completions.

Why Use This

• Better agentic work — Claude learns how to decompose complex tasks, dispatch specialized subagents, validate output quality, and self-correct through revision loops
• Agent reliability — Completion verification catches premature "done" claims, hallucination detection flags fabricated content, context health monitoring prevents session degradation, and constitutional safety rules block dangerous actions
• Multi-session execution — The execution harness breaks large projects into checkpointed tasks that survive session boundaries, with resume capability and human review gates
• Prompt quality enforcement — 35 anti-pattern checks catch vague, unsafe, or wasteful prompts before they run
• Project structure — Auto-scaffolds any new project with a self-documenting 4-layer structure and modular .claude/ configuration
• Adaptive behavior — Agent identity learns your preferences over time, skill memory accumulates domain knowledge from past tasks, compute adaptation degrades gracefully under resource pressure
• Domain expertise on demand — From double-entry accounting to circadian lighting to 3D print optimization, Claude gets expert-level methodology loaded exactly when needed
• More autonomous sessions — Skills include permission patterns, hook configurations, and structured workflows that let Claude work more independently with fewer interruptions

Installation

Three commands. Once installed, skills and permissions are set up across all projects forever.

In Claude Code, run:

/plugin marketplace add Kgan01/ghengis-skills
/plugin install ghengis-skills@ghengis-skills-marketplace
/ghengis-skills:setup

The first two install the plugin. The third configures autonomous permissions — safe dev tools are auto-allowed, dangerous operations are blocked. Restart Claude Code after setup.

All 46 skills are now available in every session — CLI, desktop app, and mobile. Claude loads them automatically when it detects a matching task.

For Teams

Add the marketplace to your project's .claude/settings.json so teammates get it automatically:

{
  "extraKnownMarketplaces": {
    "ghengis-skills-marketplace": {
      "source": {
        "source": "github",
        "repo": "Kgan01/ghengis-skills"
      }
    }
  },
  "enabledPlugins": {
    "ghengis-skills@ghengis-skills-marketplace": true
  }
}

Autonomous Permissions (Recommended)