study8677/architecture-copilot
Productivity workflowThis page was translated automatically. Technical details may differ from the original. View the original
🧭 The 'co-pilot' skill for Claude Code / Cursor / Codex — guiding you to think carefully about the architecture before writing code with ongoing in-depth questions when starting new projects. Knowledge sourced from awesome-architecture. Bilingual in Chinese and English.
Overview
A 'co-creation' skill for use. Before you write the first line of code, it uses ** to help you think through the system architecture—not to decide for you, but to help you see the trade-offs behind every choice. AI can already write most of your code for you. But there is one thing it won't do proactively for you, and it is precisely the key to the project's success or failure: Most AI programming assistants default to "." So often, without thinking about boundaries, scale, consistency, or trade-offs, you get dragged into implementation details — by the time architectural pitfalls are exposed, it's already months later and it's time to start over. : It transforms AI from a "coding intern eager to write" into a "senior architect who helps you think things through first." You say you want to make something, but it doesn't draw diagrams or code; instead, it ** until the architecture emerges on its own in conversation. It follows a fixed guided process (adapted from awesome-architecture's "Soul Six Questions" and "Eight-Step Design Method"): One is a "teaching material," the other is a "personal trainer who can ask questions." When guiding you, the co-pilot will use the templates and methodologies from the former (for example, if you're in e-commerce, it will bring up the key decisions of the e-commerce platform's template to question you). Feel free to add more system types of 'critical decision questions,' improve guiding scripts, or translations. First change skills/architecture-copilot/SKILL.md (the authoritative version), then synchronize .cursor/rules/architecture-copilot.mdc with AGENTS.md; If there are tutorials/templates, also sync README.md, README_en.md, INSTALL.md.
Overview (machine translated)
A ‘co-creation’ skill for Claude Code / Cursor / OpenAI Codex. Before you write the first line of code, it uses continuous deep questions to help you think through the system architecture—not for you to decide, but to help you see the trade-offs behind each choice.
English · Chinese Simplified
Why is it needed?
AI can already write most of your code for you. But there is one thing it won’t do proactively for you, and it is precisely the key to the project’s success or failure:
**Before you start, think carefully about what this system should look like. **
Most AI programming assistants’ default behavior is, “You say what you want, I’ll start writing immediately.” So often, without clearly considering boundaries, scale, consistency, or trade-offs, you get led into implementation details by the time architectural pitfalls are exposed, months later, it’s time to start over.
Architect Co-pilot Goes Against the Grain: It transforms AI from a ‘coding intern’ into a ‘senior architect who helps you think things through first.’ If you say you want to build something, it doesn’t draw diagrams or code, but instead asks question after question until the architecture emerges in conversation.
How it works
It follows a fixed guided process (adapted from awesome-architecture’s “Soul Six Questions” and “Eight-Step Design Method”):
Ultimately, it helps you converge on: Architecture Panorama (ASCII), data models, a string of ADR decision records, scalable bottleneck warnings, MVPs→ mature evolution roads, and an honest risk list.
Seven stages
| Stage | What is it doing |
|---|---|
| 0 · Opening | Just one question: What do you want to make? Which product do you resemble the most? |
| 1 · Business and Scope | Who is it solving what problems it is for? More importantly, this edition does not do anything |
| 2 · Soul Six Questions | Scale / Read-write ratio / Consistency / Growth / Cost of failure / Constraints |
| 3 · Envelope Back Estimation | Calculate QPS and storage capacity on the spot, and judge ‘What will this system be overwhelmed?’ |
| 4 · Quality attribute trade-offs | Performance/availability/consistency/cost… Passing item by item, forcing you to rank, it’s impossible to have everything |
| 5 · Key Decision Follow-up ⭐ | By system type, present fork points one by one: Option A vs B, what is the cost for each? |
| 6 · Convergence output | Architecture diagram + data model + ADR + bottleneck + evolution roadmap + risk |
| 7 · Counter-challenge | Proactively pointing out: Where will it die? What have you given up? |
Strict rule throughout: **Ask before answer, one dimension at a time, every choice is followed by ‘why/cost’, avoid getting stuck in grammar, and do your best to subtract. **
🛠 Three tools, ready to use immediately
| Tools | Shape | Where to place |
|---|---|---|
| Claude Code | skills/architecture-copilot/SKILL.md |
~/.claude/skills/ or the item .claude/skills/ |
This page was translated automatically. Technical details may differ from the original.
Recommended Tools
Try a different keyword or remove a filter.
Install
npx skillfish add study8677/architecture-copilot