This project is a practical manual and starter kit for running a complete workflow with AI, from product requirements and implementation planning to coding, verification, and code review.
概览
This project is a practical manual and starter kit for running a complete workflow with AI, from product requirements and implementation planning to coding, verification, and code review. It provides a skills-first, reusable structure you can copy into your own repository to operationalize SDD in day-to-day delivery. It is useful because it turns SDD from a high-level idea into a repeatable system with shared standards, canonical prompts, and portable conventions that stay consistent across Codex, Cursor, and Claude. It is highly recommended to use it with a spec-driven process such as OpenSpec. If you want to try our best-practices in an Openspec-ready ecosystem, check out our Openspec AI Specs alternative This repository keeps a single canonical source in ai-specs/ and exposes it to each copilot folder using symlinks: - .codex/* links to canonical resources - .cursor/* links to canonical resources - .
README
以下为原文 README 节选,完整内容请查看原文。
AI Specs for Skills-First Development
This project is a practical manual and starter kit for running a complete Spec-Driven Development (SDD) workflow with AI, from product requirements and implementation planning to coding, verification, and code review. It provides a skills-first, reusable structure you can copy into your own repository to operationalize SDD in day-to-day delivery.
It is useful because it turns SDD from a high-level idea into a repeatable system with shared standards, canonical prompts, and portable conventions that stay consistent across Codex, Cursor, and Claude.
It is highly recommended to use it with a spec-driven process such as OpenSpec.
If you want to try our best-practices in an Openspec-ready ecosystem, check out our Openspec AI Specs alternative
Repository Structure
.
├── ai-specs/
│ ├── .agents/ # Canonical agent role definitions
│ ├── .commands/ # Small set of shared utility commands
│ └── skills/ # Canonical skill definitions (main workflow entrypoint)
│
├── .codex/
│ ├── agents -> ../ai-specs/.agents
│ ├── commands -> ../ai-specs/.commands
│ └── skills -> ../ai-specs/skills
│
├── .cursor/
│ ├── agents -> ../ai-specs/.agents
│ ├── commands -> ../ai-specs/.commands
│ ├── skills -> ../ai-specs/skills
│ └── rules/
│
├── .claude/
│ ├── agents -> ../ai-specs/.agents
│ ├── commands -> ../ai-specs/.commands
│ └── skills -> ../ai-specs/skills
│
├── docs/ # Project technical context and reference docs
└── README.md
Multi-Copilot Strategy
This repository keeps a single canonical source in ai-specs/ and exposes it to each copilot folder using symlinks:
.codex/*links to canonical resources.cursor/*links to canonical resources.claude/*links to canonical resources
Why This Approach
- Single source of truth: one canonical definition for agents, commands, and skills
- No duplicated maintenance: update once, all copilot folders stay aligned
- Tool compatibility: each copilot reads from its expected folder structure
- Safe evolution: workflows can change without reorganizing every tool-specific folder
Skills-First Workflow
Use skills as the default entrypoint for recurring tasks.
Current examples in this repository:
ai-specs/skills/enrich-user-story/SKILL.mdai-specs/skills/write-pr-report/SKILL.md
Commands still exist as lightweight utilities in ai-specs/.commands, but the main functional workflows should be implemented as skills.
Technical Context Location
Project-level technical context now belongs in docs/, for example:
docs/doc_architecture.mddocs/doc_ai_planning_mode.mddocs/doc_verification_guide.md
If you bootstrap this setup into another project, replace these documents with your own architecture, planning, and verification references.
Quick Start
推荐工具
换一个关键词,或者移除筛选条件。
安装
npx skillfish add ncltrdorsor/manual-sdd