ワンクリックで
agents-md
AGENTS.md authoring patterns. Activated when writing AI agent context files, working with agents.md or AGENTS.md.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
メニュー
AGENTS.md authoring patterns. Activated when writing AI agent context files, working with agents.md or AGENTS.md.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
Codebase onboarding guide. Activated when starting a new project analysis, requesting systematic codebase understanding, or running /onboard.
Feature-Sliced Design architecture patterns. Activated when working with FSD layers, slices, dependency rules, or structuring frontend code.
Modern React 19+ patterns. Activated when working with state management, Suspense, Compound Components, or React hooks.
React Query server state management patterns. Activated when applying Query Factory, queryOptions, or mutationOptions patterns.
Practical TypeScript patterns. Activated when working with type inference, utility types, generics, type guards, or type narrowing.
Writing readable, maintainable code. Activated when applying naming conventions, function decomposition, or Early Return patterns.
SOC 職業分類に基づく
| name | agents-md |
| description | AGENTS.md authoring patterns. Activated when writing AI agent context files, working with agents.md or AGENTS.md. |
Vercel research: Skills 53% → AGENTS.md 100% pass rate. Passive context beats on-demand skill invocation.
| Approach | Reliability | Reason |
|---|---|---|
| Skills | ~53% invocation | Trigger instability, wording-sensitive |
| Explicit skill hints | ~95% | Still fragile to prompt changes |
| AGENTS.md | 100% | Always loaded in context |
# Project Name
## Tech Stack
React 19, Next.js 15 (App Router), TypeScript 5.7, Tailwind CSS 4
## Architecture
- FSD (Feature-Sliced Design)
- Server Components default, 'use client' minimal
## Conventions
- Components: PascalCase, files: kebab-case
- barrel export: entities/features index.ts only
## Patterns
|state: react-query (server), zustand (client)
|forms: @tanstack/form + zod standard schema
## Commands
|dev: bun dev
|build: bun run build
|test: bun run test
## DO NOT
- any type
- console.log in commits
- useEffect for data fetching (use react-query)
40KB → 8KB (80% compression) with same performance. Provide an index map for agent retrieval.
[Docs Index]|root: ./.project-docs
|IMPORTANT: Prefer retrieval-led reasoning over pre-training-led reasoning
|api/:{auth.md,users.md,payments.md}
|patterns/:{state-management.md,data-fetching.md,error-handling.md}
|guides/:{setup.md,deployment.md,testing.md}
project-root/
├── AGENTS.md # Global conventions, tech stack
├── apps/
│ ├── web/
│ │ └── AGENTS.md # Web-specific rules
│ └── mobile/
│ └── AGENTS.md # Mobile-specific rules
├── packages/
│ └── ui/
│ └── AGENTS.md # UI library rules
└── .project-docs/ # Reference docs directory
## IMPORTANT
- Check .project-docs/ before using framework APIs
- Prioritize project docs over pre-trained knowledge
- Read docs first when uncertain, then write code
# BEFORE (verbose) - 200 tokens
## State Management
We use React Query for server state management. All queries should use
the queryOptions pattern defined in the entities layer.
# AFTER (compressed) - 50 tokens
## State
|server: react-query (queryOptions pattern, entities layer)
|client: zustand stores
## API Patterns
|IMPORTANT: Read .project-docs/api-patterns.md when writing APIs
|summary: queryOptions + mutationOptions, Key Factory required
|details: ./.project-docs/api-patterns.md
## DO NOT
- any type
- console.log in commits
- re-export internals from index.ts (public API only)
- useEffect for data fetching (use react-query)
Negative instructions effectively prevent common agent mistakes.
For Next.js projects:
npx @next/codemod@canary agents-md
This detects framework version, downloads matching docs, and injects compressed index into AGENTS.md.