---

What It Does

Before: Your AI agent gives generic advice because it doesn't know your project. Different stack, same suggestions.

After: Your agent has an AGENTS.md that matches what you actually use — every command, convention, and architecture decision derived from your real package.json, config files, and directory structure.

npx skills add OJPalenzuela/agents-generator

Then in any project:

create AGENTS.md

What You Get

From a Next.js + Bun + Tailwind + Vitest project, the skill produces:

AGENTS.md                          ← Commands, conventions, verification cycle, PR rules
.agents/rules/
  architecture.md                  ← Stack diagram, routes, data flow
  frontend-patterns.md             ← Components, state, trust boundaries
  server-actions.md                ← Entry points, error handling, rate limiting
  testing.md                       ← "74 tests in 5 files", vitest commands
  git-workflow.md                  ← Conventional commits, pre-commit checks

Rules that don't apply (backend without NestJS, database without ORM) are skipped automatically.

The generated AGENTS.md opens with real content, not placeholders:

## Setup commands
- Install deps: `bun install`
- Start dev: `bun dev`
- Run tests: `bun run test:run`

## Essential Commands
| bun dev          | Next.js dev server  |
| bun run test:run | Vitest single-run   |
| bun doctor       | react-doctor check  |

## Verification Cycle
bunx tsc --noEmit → bun run lint → bun run test:run → bun doctor

Minimal mode — single file following the agents.md standard:

## Setup commands
- Install deps: `pnpm install` · Run tests: `pnpm test`

## Code style
- TypeScript strict mode · No `any` types · UI in Spanish, code in English

## Testing instructions
- Run `pnpm test` before merging · Add tests for code you change

## PR instructions
- Conventional commits · No AI attribution in commits

Monorepo Support

Detects workspaces and generates per-package AGENTS.md files:

packages/db/AGENTS.md        ← Prisma client, build command, entrypoints
packages/validation/AGENTS.md ← Zod schemas, exports map
apps/dashboard/AGENTS.md     ← Next.js, vitest, port 3000

Modes

Mode	Command	Output
Full (default)	create AGENTS.md	Rich AGENTS.md + 6-10 companion rule files
Minimal	create a simple AGENTS.md	Single ~30-line AGENTS.md (agents.md standard)
Update	update AGENTS.md	Diff existing, regenerate only what changed
Dry-run	preview AGENTS.md	Show what would be generated, write nothing

Usage

Install:

npx skills add OJPalenzuela/agents-generator

Or manually:

git clone https://github.com/OJPalenzuela/agents-generator ~/.config/opencode/skills/agents-generator

Run from any project:

create AGENTS.md              # Full mode
create a simple AGENTS.md     # Minimal mode
update AGENTS.md              # Update existing
preview AGENTS.md             # Dry-run

After generation, review your setup:

• Commit .agents/rules/ — they're part of your project documentation
• Add .agents/backups/ to .gitignore — backups don't need to be committed
• Run the verification cycle once to confirm every command works

Detection

Reads your project and detects 16 categories automatically:

Category	Examples
Package manager	bun, pnpm, npm, yarn
Framework + router	Next.js (App/Pages Router), NestJS, Vite, Astro, Express
CSS	Tailwind, shadcn/ui, CSS Modules, styled-components
Testing	Vitest, Jest, Playwright, Cypress
ORM + provider	Prisma (postgres/sqlite), Drizzle, Knex, Mongoose
Validation	Zod, Yup, class-validator, plain TypeScript
State	Zustand, Redux, Jotai, React Query, SWR
API pattern	Server Actions, tRPC, GraphQL, REST
Forms, i18n, Auth	react-hook-form, next-intl, NextAuth, Clerk
Monorepo	pnpm workspaces, Turborepo, Nx, Lerna

Plus: design token extraction, @import for existing docs, confidence scoring, managed blocks, multi-platform output.

Quality Guarantees

• Validates commands — every command verified against package.json scripts before writing
• No placeholders — output scanned for {{, TODO, ... — rejected if any remain
• Backup before overwrite — existing files copied to .agents/backups/ with timestamp
• Line limits — 300-line soft cap, 500-line hard cap
• agents.md compatible — follows the standard used by 60k+ open-source projects

License

MIT — see LICENSE