system-wrightlisted
Install: claude install-skill EthanQC/SystemWright
# SystemWright
## Core Rule
This skill itself is the product layer for idea refinement and AI work-system design.
Do not treat MCP, orchestrators, tools, dashboards, or new agents as prerequisites for this skill. They are possible outputs that may be designed after the user's goal is understood and confirmed.
Always separate two layers:
1. Layer 1: refine the user's vague idea into a confirmed work target.
2. Layer 2: design the best-fit work system using Prompt, Context, Harness, Loop, and human judgment.
If the user is non-technical, explain every design choice in ordinary work language before naming the technical pattern.
## Required Workflow
### 0. Route the Request
Use this skill when the user asks for:
- turning a vague idea into an AI workflow or AI operating system
- improving a recurring business process with AI
- deciding whether something should become a prompt, skill, MCP, automation, loop, or orchestrator
- applying Prompt, Context, Harness, and Loop to a real project
- building a reusable way to work with an agent CLI or agentic IDE (Codex, Claude Code, Gemini CLI, Cursor, or compatible)
If the user only needs a one-off answer, do not over-design. Deliver the concrete thing they asked for first (the summary, the speech, the checklist), then offer in one optional line to systematize it later. Do not produce a Work System Card, MCP decision, or Trial Run for a one-off request.
Named-invocation override: if the user invokes this skill by name (for examp