How newtype-01/newtype-skills works
Project type & purpose (what you’re diagramming)
This repo is not an executable software system (no runtime services, no code, no build/deploy). It’s a content-based prompt/skill library: a collection of methodology-driven prompt “skills” (Markdown documents) meant to be copied/pasted or loaded by AI tooling (OpenCode / Claude Code / Custom GPT / Claude Project).
• Artifacts: the skills as documents (SKILL.md files)
• Consumers: humans and AI assistants
• Integration points: tools that can load skills from a directory (e.g., ~/.claude/skills, .opencode/skill) and “/skill” commands
• Flows: how a user selects/installs/loads a skill and how it influences AI outputs
File structure analysis (map repo structure into components)
Top-level:
• README.md, README.zh-cn.md: entrypoints / documentation / usage instructions
• skills/: the core library namespace
• skills/<skill-name>/SKILL.md: each skill is a self-contained prompt methodology document
Concrete skill modules:
• skills/super-analyst/SKILL.md
• skills/super-writer/SKILL.md
• skills/super-fact-checker/SKILL.md
• skills/super-editor/SKILL.md
• skills/super-interviewer/SKILL.md
Architecturally this is a modular catalog: each folder is a module; skills/ is a package index; README is the “registry UI” (human-facing).
README-derived architecture cues (how it’s used)
The README describes three primary consumption modes:
Direct Copy-Paste into a conversation with *any* AI assistant
2. Tooling integration (OpenCode / Claude Code) by cloning into a specific skills directory and loading via /skill <name>
Embedding into Custom GPT / Claude Project system prompts (i.e., saved configuration)
Key non-functional attributes to reflect:
• Tool-agnostic: works across ChatGPT/Claude/Gemini etc.
• Language: skills written primarily in Simplified Chinese but methodologies are universal
• License: MIT (important for distribution/reuse)
Repository (this project)
• README.md / README.zh-cn.md (documentation)
• skills/ (skill catalog)
• super-analyst/SKILL.md
• super-writer/SKILL.md
• super-fact-checker/SKILL.md
• super-editor/SKILL.md
• super-interviewer/SKILL.md
Distribution / Installation targets
• GitHub (source of truth)
• Local filesystem clone (e.g., ~/.claude/skills/huangyihe-skills or .opencode/skill/huangyihe-skills)
Skill loader environments
• OpenCode
• Claude Code
• (Optional) other AI tools that support skill loading / prompt presets
Execution / Runtime environment (conceptual)
• AI assistant / LLM (ChatGPT, Claude, Gemini…)
• Conversation context (where the skill prompt is injected)
Actors
• User (selects skill, loads it, asks tasks)
• (Optional) Contributor (adds/edits SKILL.md; PR flow)
Architectural patterns/design principles
• Modular skill packaging: one skill per folder, single entry file SKILL.md
• Tool-agnostic prompt interface: “prompt text” is the integration contract
• Registry/index pattern: README is the catalog; skills/ is the namespace
• Immutable-ish artifacts: skills are versioned via Git; distribution via clone/fork
Relevant technologies/frameworks (even if “soft”)
Since there’s no runtime stack, your “tech” labels should reflect:
• Markdown (.md) as artifact format
• Git/GitHub as distribution/versioning mechanism
• OpenCode / Claude Code skill loader (integration mechanism)
• AI assistant platforms (execution targets)
3. Component view (C4 Level 3): show each skill as a component with its purpose (Analyst/Writer/Fact-checker/Editor/Interviewer)