AI Workflow

The boilerplate ships portable agent guidance so the same project conventions work across Claude Code, Codex, Cursor, Amp, and Jules.

What's included

AGENTS.md guides

The primary guidance lives in nested AGENTS.md files:

  • AGENTS.md at the repo root explains the two main surfaces in the monorepo, top-level commands, and shared coding and UI conventions.
  • apps/boilerplate/AGENTS.md is the nested guide for the SaaS starter and covers auth, database, billing, email, AI chat, SEO, and hardening rules.

AI Coding Skills (skills/)

Portable skills live under apps/boilerplate/skills/ and encode repeatable workflows for common tasks:

  • add-protected-page: Adds a protected /app/* route and updates navigation.
  • add-model: Adds a Drizzle schema, types, and query helpers.
  • add-plan: Adds a billing plan across pricing config, env vars, and provider mappings.
  • add-email-template: Adds a new transactional email template and sender helper.
  • generate-env-secret: Generates a secure random secret for auth and similar env vars.
  • wire-email-to-event: Connects an email helper to the correct central flow.
  • wire-magic-link-auth: Audits or customizes the magic-link auth flow.
  • seo-audit-page: Audits metadata, sitemap coverage, and structured data for a route.
  • security-audit: Reviews auth, injections, rate limiting, logging, headers, secrets, and webhooks.
  • generate-legal: Generates privacy, terms, and imprint pages.
  • deslop: Removes AI-generated slop from the current diff.

Installing the skills

Install the portable skills into compatible tooling with:

Code
npx skills add . -y

Skills are defined as one folder per task with a SKILL.md entrypoint. For example:

Code
apps/boilerplate/skills/
  add-model/SKILL.md
  security-audit/SKILL.md
  wire-magic-link-auth/SKILL.md

How AGENTS and skills work together

  • AGENTS.md carries durable repo context and conventions.
  • SKILL.md files carry reusable workflows for repeatable tasks.
  • The nearest AGENTS.md wins, so work inside apps/boilerplate should follow the nested boilerplate guide first.

The boilerplate still ships .cursor/rules/ and .cursor/commands/ for Cursor users, but those files are now the compatibility layer rather than the primary interface.

Best practices

  • Keep AGENTS.md files focused on conventions, boundaries, and commands.
  • Keep skills focused on repeatable workflows with inputs, steps, and verification.
  • Update both when the boilerplate architecture changes.
  • Preserve .cursor/ assets when you need Cursor compatibility, but keep new guidance portable by default.