skills/ai-instruction-standards/SKILL.md
Create and maintain AI instruction files (CLAUDE.md, AGENTS.md, .cursor/rules/, etc.) with proper structure. Use when: creating AI instruction files, separating universal vs project-specific rules, configuring AI tools. Keywords: CLAUDE.md, AGENTS.md, cursorrules, windsurfrules, clinerules, AI instructions, system prompt.
npx skillsauth add asiaostrich/universal-dev-standards ai-instruction-standardsInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Language: English | 繁體中文
Version: 2.0.0 Last Updated: 2026-04-28 Applicability: All AI coding tools
Core Standard: This skill implements AI Instruction File Standards. For comprehensive methodology documentation, refer to the core standard.
This skill is part of a three-layer AI collaboration system:
| Layer | Skill | Question it Answers |
|-------|-------|-------------------|
| Behavior (Immediate) | /ai-collaboration | "How should AI respond accurately?" |
| Configuration (Session) | /ai-instruction-standards (this) | "What to write in CLAUDE.md / AGENTS.md?" |
| Architecture (Long-term) | /ai-friendly-architecture | "How to structure code for AI?" |
This skill helps create and maintain AI instruction files with proper separation between universal standards and project-specific configurations, across all major AI coding tools.
| Tool | Primary File | Workflow Mechanism | MCP |
|------|-------------|-------------------|-----|
| Claude Code | CLAUDE.md + .claude/rules/*.md | Skills (.claude/skills/ → /{name}) | ✅ |
| Gemini CLI | GEMINI.md | .gemini/commands/*.toml → /{name} | ✅ |
| OpenAI Codex CLI | AGENTS.md (+ AGENTS.override.md) | Team commands; /review built-in | ✅ |
| OpenCode | AGENTS.md (CLAUDE.md compatible) | Built-in only (/init /undo /share) | ✅ |
| Tool | Primary File | Workflow Mechanism | MCP |
|------|-------------|-------------------|-----|
| Cursor | .cursor/rules/*.mdc ⚠️ | @-mentions; /multitask | ✅ |
| GitHub Copilot | .github/copilot-instructions.md | .github/prompts/*.prompt.md → /{name} | ✅ |
| Windsurf | .windsurfrules / .windsurf/rules/*.md | .windsurf/workflows/*.md → /{name} | ✅ |
| Cline | .clinerules | None | ✅ |
⚠️ Cursor:
.cursorrulesis deprecated — migrate to.cursor/rules/*.mdc
AGENTS.mdAGENTS.md is the emerging de-facto cross-tool instruction standard:
Supported by: Gemini CLI, OpenAI Codex CLI, OpenCode, GitHub Copilot, Windsurf, Cursor
Not supported by: Claude Code (uses CLAUDE.md), Cline (uses .clinerules)
Recommendation: Use AGENTS.md as the universal baseline for cross-tool projects, then add tool-specific files for advanced features (Skills, Workflows, Prompts).
| Type | Contains | Example |
|------|----------|---------|
| Universal | Generic rules | "Run tests before committing" |
| Project-Specific | Concrete commands | "Run npm test before committing" |
# [Project Name] - AI Instructions
## Universal Standards
<!-- Rules applicable to ANY project -->
- Commit message format
- Code review checklist
- Testing standards
- Anti-hallucination rules
---
## Project-Specific Configuration
<!-- Unique to THIS project -->
### Tech Stack
[Your technologies here]
### Quick Commands
[Your build/test/deploy commands]
### File Structure
[Your project structure]
CLAUDE.md # Main instructions (hierarchical: global → project → subdir)
.claude/rules/ # Glob-scoped additional rules
.claude/skills/{name}/SKILL.md # Custom slash commands → /{name}
.claude/agents/{name}.md # Subagent definitions
GEMINI.md # Main instructions
.gemini/commands/{name}.toml # Custom slash commands → /{name}
.gemini/agents/{name}.yaml # Subagent definitions
Example .gemini/commands/review.toml:
description = "Run code review checklist"
prompt = "Review the following changes: !{git diff HEAD}"
AGENTS.md # Main instructions (Git root → cwd traversal)
AGENTS.override.md # Temporary override (highest priority)
~/.codex/AGENTS.md # Global fallback
.codex/agents/ # Custom agent definitions
AGENTS.md # Primary (auto-recognized)
CLAUDE.md # Also recognized (migration compatibility)
.opencode/agents/ # Custom agent definitions
opencode.json (instructions) # Glob-pattern file references
.cursor/rules/ # MDC format rules (replaces .cursorrules)
{name}.mdc # Frontmatter: description, globs, alwaysApply
AGENTS.md # Also supported for agent context
MDC frontmatter example:
---
description: "TypeScript coding standards"
globs: ["**/*.ts", "**/*.tsx"]
alwaysApply: false
---
Migration: If you have
.cursorrules, move content to.cursor/rules/*.mdc.
.github/copilot-instructions.md # Always-on, all chats
.github/instructions/*.instructions.md # File-glob scoped (applyTo frontmatter)
.github/prompts/*.prompt.md # Reusable templates → /{name} slash commands
.github/agents/*.agent.md # Custom agents with tool access control
AGENTS.md # Also recognized
.windsurfrules # Project rules (team-shareable)
.windsurf/rules/*.md # MDC frontmatter structured rules
.windsurf/workflows/*.md # Reusable task sequences → /{name}
AGENTS.md # Also recognized
Workflow example (.windsurf/workflows/review.md):
Run a code review:
1. Run `git diff HEAD`
2. Check for BLOCKING issues (security, correctness)
3. Check for IMPORTANT issues (design, tests)
4. Output findings with BLOCKING/IMPORTANT/SUGGESTION prefixes
When a project uses multiple AI tools:
project/
├── AGENTS.md # Universal baseline (cross-tool)
├── CLAUDE.md # Claude Code (extends AGENTS.md)
├── GEMINI.md # Gemini CLI
├── .cursor/rules/
│ └── standards.mdc # Cursor
├── .windsurf/
│ └── workflows/ # Windsurf workflows
│ ├── review.md
│ └── checkin.md
└── .github/
├── copilot-instructions.md # Copilot always-on
└── prompts/
└── review.prompt.md # Copilot slash command
Best Practice: Write universal content in AGENTS.md once, then import/reference it from tool-specific files to avoid duplication.
| Category | Good Examples | |----------|---------------| | Commit Standards | "Follow Conventional Commits format" | | Code Review | "Use BLOCKING, IMPORTANT, SUGGESTION prefixes" | | Testing | "Maintain 80% coverage minimum" | | AI Behavior | "Always read code before analyzing" |
Avoid in Universal Sections:
npm test, pytest)cli/src/, /var/www/)Node.js 18, Python 3.11)| Category | Examples |
|----------|----------|
| Tech Stack | Node.js 18, React 18, PostgreSQL 15 |
| Commands | npm run lint, ./scripts/deploy.sh |
| File Structure | src/, cli/, tests/ |
| Team Conventions | Traditional Chinese comments |
Before committing changes to AI instruction files:
.cursorrules migrated to .cursor/rules/*.mdcAGENTS.md covers the universal baselineCLAUDE.md, AGENTS.md, GEMINI.md, or equivalent files.cursor/, .windsurf/, .github/copilot-instructions.md, etc.)If no AI instruction file found:
AGENTS.md for cross-tool projects, CLAUDE.md for Claude Code only.gitignore if contains sensitive infoAfter /ai-instructions completes, suggest:
- Create or update project's
CLAUDE.md/AGENTS.md⭐ Recommended — Apply standards immediately- Run
/ai-friendly-architectureto optimize AI collaboration at the architecture level- Run
/ai-collaborationto review AI behavior guidelines
| Version | Date | Changes | |---------|------|---------| | 2.0.0 | 2026-04-28 | Add Gemini CLI, OpenAI Codex CLI; update Cursor (MDC format, deprecated .cursorrules); update OpenCode (AGENTS.md primary); update Copilot (multiple file types); update Windsurf (Workflows); add AGENTS.md cross-tool standard section | | 1.0.0 | 2026-01-25 | Initial release |
This skill is released under CC BY 4.0.
Source: universal-dev-standards
development
[UDS] 扫描代码库的调试残留与代码质量问题;可自动修正安全模式。 Use when: before committing, during PR review, or periodic codebase cleanup. Keywords: sweep, debug cleanup, console.log, debugger, TODO, ts-any, code quality, 扫描, 清理.
tools
[UDS] 从规格衍生 BDD 场景、TDD 骨架或 ATDD 表格
development
[UDS] 识别重复流程并以正确的开发深度构建 Skill
tools
[UDS] AI 辅助 git push 安全层:质量门禁 + 协作护栏。 Use when: pushing commits, force pushing, pushing to protected branches, pushing feature branches. Keywords: git push, force push, protected branch, quality gate, push receipt, PR automation, 推送, 保护分支, 质量门禁.