A minimal, opinionated starter template for vibe-coding and AI-agent-driven development. Use it to bootstrap new projects with a stable docs/-first workflow so your coding agent always has the right context.
Solo developers who want to ship faster with AI coding agents, without losing track of scope, architecture, or past decisions.
AI coding agents work best when they have a clear blueprint. This template separates thinking and guardrails from source code, so your agent reads intent first and stays aligned across sessions.
docs/: the core docs layerprd.md: product requirements and scopearchitecture.md: system architecture and tech stackrules.md: coding rules and agent guardrailsdesign.md: UI/UX design tokens and decisionstasks.md: executable task checklistmemory.md: lessons learned and session memory
AGENTS.md: canonical agent instructions (readdocs/first, verify before ticking tasks, stop-and-ask gates). Read natively by Cursor, OpenCode, Codex, Kilo and others; Claude Code loads it viaCLAUDE.mdCLAUDE.md: one-line import ofAGENTS.mdfor Claude Code.cursorrules: legacy Cursor pointer toAGENTS.mdexamples/tinytodo/: a fully filled-indocs/set showing what good looks like.github/ISSUE_TEMPLATE: standard feedback issue template.gitignore: sensible defaults for common stacksLICENSE: MIT license
The diagram shows the documented operating model. Enforcement depends on your agent and tooling; the template itself does not block agents.
flowchart LR
A[Use this template] --> B[Edit docs/ for your project]
B --> C[Open in Cursor / Claude Code]
C --> D{Agent reads docs/ first}
D --> E[Build from tasks.md]
E --> F[Record lessons in memory.md]
F --> D
sequenceDiagram
actor Dev as You
participant Agent as Coding agent
participant Docs as docs folder
participant Code as Source code
Dev->>Docs: Define PRD, architecture, rules, design
Dev->>Agent: Start a session
Agent->>Docs: Read intent and guardrails
Agent->>Docs: Pick next item in tasks.md
Agent->>Code: Implement the task
Agent->>Docs: Update tasks.md and memory.md
Agent-->>Dev: Summary and open questions
Your docs/ folder is the source of intent. The agent reads it before writing code, so scope, architecture, and rules stay consistent across sessions.
Every core doc (prd.md, architecture.md, design.md, rules.md) starts with Status: draft. Agents are instructed not to write code until you flip each to ready. The template gives agents this rule in writing; it does not enforce it with tooling.
- Click Use this template on GitHub, or scaffold from the terminal (works on Windows, macOS, Linux):
npx degit abhilash333naidu/AI_Project_Setup my-ai-project
cd my-ai-project
git initdegit copies the files without the template's git history. No rm -rf .git needed.
- Edit each file in
docs/for your project. - Set
Status: readyat the top ofprd.md,architecture.md,design.mdandrules.mdonce filled. - Open the project in Cursor, Claude Code, or your preferred agent host.
- Start coding. The agent reads
AGENTS.mdanddocs/first, verifies each task before ticking it, and keepstasks.mdandmemory.mdcurrent.
Download AI_Project_Setup_template.zip from the v1.2.0 release, or click Use this template above.
Open an issue using the feedback template. Useful feedback includes:
- What was missing when you started using it
- Which docs file was unclear
- What automation would have helped
Pull requests are welcome. Read CONTRIBUTING.md first. Release history is in CHANGELOG.md. To report a security issue, follow SECURITY.md and do not open a public issue.
MIT. See LICENSE.
