Skip to content

nghyane/agent-stack

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

agent-stack

License: MIT Go Svelte agent-native

One prompt. One minute. A complete Go + SvelteKit project your AI agent understands from line zero.

Use case

You start a new full-stack project. You want a Go API and a SvelteKit frontend that talk through a typed OpenAPI contract. You want your AI coding agent to navigate it correctly from day one — following conventions, respecting architecture, never guessing.

You don't want boilerplate generators that dump code you can't explain. You don't want to manually rename 30 files and rewrite every import. You don't want scripts you can't trust.

agent-stack solves this. It ships patterns the agent reads and executes — not opaque scripts, not stale templates. Clone, bootstrap, interview. Done.

How it works

One prompt  →  agent clones  →  renames everything  →  interviews you  →  writes AGENTS.md  →  ready

1. Clone. git clone --depth 1 the repo into your project directory.

2. Bootstrap. The agent follows the BOOTSTRAP checklist in AGENTS.md: renames directories, rewrites the Go module path and every import, sets the app name and API title, repoints config — no scripts, no black boxes. The agent owns every step.

3. Interview. The agent writes the project's AGENTS.md from scratch: purpose, constraints, workflow. As the project grows, the agent adds to Decisions and Patterns. The AGENTS.md becomes a living knowledge base.

4. Ready. go build passes. pnpm run check has zero errors. Your agent knows the architecture and conventions. Start building.

Why this over alternatives

No scripts. The agent reads AGENTS.md and acts. Every rename, every replace is visible and verifiable. If something breaks, the agent knows exactly where.

No stale docs. Documents record principles and where to look — never version numbers, ports, or commands. The agent reads source for volatile details.

No copy-paste. Features are vertical slices (handler → service → data). The agent learns the pattern from an example and reproduces it for new domains — never clones a folder full of irrelevant logic.

No setup drift. Per-directory AGENTS.md files auto-load when the agent works in each directory. api/AGENTS.md for backend conventions, app/AGENTS.md for frontend. The Svelte MCP server provides live docs and static analysis.

Stack

  • Go backend — Huma (OpenAPI 3.1), sqlc (typed SQL), pgx + Postgres, goose migrations, scs sessions. Two-tier config: static env for secrets, DB-backed dynamic settings with Postgres LISTEN/NOTIFY.
  • SvelteKit frontend — Svelte 5 runes, Cloudflare Pages. Typed API client from OpenAPI spec. Runtime backend URL via $env/dynamic/public.
  • Agent tooling — Svelte MCP server in opencode.json, per-directory AGENTS.md files the agent auto-loads.

Quick start

Start a new project from https://github.com/nghyane/agent-stack:

1. git clone --depth 1 the repo into a fresh directory
2. Ask me for the project name and Go module path
3. Follow the BOOTSTRAP checklist in the repo's AGENTS.md

Paste into any AI coding agent.

License

MIT — see LICENSE.

About

Full-stack starter kit for AI coding agents — Go API (Huma, sqlc, Postgres, vertical-slice) + SvelteKit frontend (Svelte 5 runes, Cloudflare Pages). Agent-native: clone, bootstrap, interview, code. No scripts.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors