The SaaS boilerplate AI coding agents actually understand
Plain PHP, jQuery and SQLite is the most training-data-saturated stack in existence. No build step, no transpiler, and the whole app fits in your agent’s context — with the agent docs already written.
An AI coding agent is good at exactly what it has seen a million times. Plain PHP, jQuery and SQLite have decades of documentation, Stack Overflow answers and working open-source code behind them — it is the stack LLMs know best, by sheer weight of training data. Simple Stack leans into that: no framework conventions to guess at, no transpiler between the code and what runs, and a codebase small enough for an agent to hold whole in context. That is what an agent-ready codebase actually means — not a marketing line, a property of the code.
- AGENTS.mdThe open agents.md standard — one file that tells any coding agent how the project is laid out and how to work in it.
- CLAUDE.mdProject instructions Claude Code picks up automatically: where logic lives, the default workflow, what not to introduce.
- AI_TASK_PROMPTS.mdReady-made prompts for common jobs — add a page, wire a webhook, change a plan — written against the real codebase.
- AI_HANDOFF.mdThe architecture handoff an agent reads before its first edit, so session two starts where session one ended.
One file per job, so edits land first try
Claude Code, Cursor and Codex do their best work when there is exactly one place to look.
Every page renders from one entry point. An agent looking for a route reads one file — no router config to trace.
All JSON endpoints in one place. Add an action and the pattern to copy is right above it.
All the jQuery in a single file — no components, no bundler, no generated output to reverse-engineer.
Built for vibe coding sessions that don’t fall apart
Long agent sessions usually die one of two ways: the toolchain breaks (a build error the agent cannot see past) or the context runs out (a framework spreads one feature across a dozen files). A flat, no-build-step codebase removes both failure modes — what the agent writes is what runs, and the whole app stays in view. The reasoning is on the no-build-step stack page; if you are shipping solo with an agent, the indie-hacker template page covers that workflow.
And the docs are deliberately tool-agnostic: AGENTS.md is an open standard, so Claude Code, Cursor and Codex all read the same instructions.
Buy it, open Claude Code, point it at the docs.
The full source and its agent docs — £49 once. Or click through the demos first.