Skip to content

Built for the AI Era

Most frameworks were designed to be maintained and consumed by humans reading tutorials. The pico ecosystem is additionally designed to be read, extended and maintained by AI coding assistants — and ships the artifacts that make that work in practice, not as marketing.

What that means, concretely

Every repo carries its own machine-readable playbook. Each pico-* module ships an AGENTS.md (project structure, commands, key concepts, code style, boundaries) and a CLAUDE.md (ecosystem context and critical invariants). An AI assistant — or a new human contributor — opens the repo already knowing where things go, how to run the tests, and which lines must never change.

Conventions are uniform across the ecosystem. Same layout, same pyproject.toml shape, same test and docs structure in every module, enforced by a shared checklist. For an AI maintainer this is the difference between reasoning once and reasoning fifteen times: a fix learned in one module applies mechanically to the rest.

Decisions are written down, with their rejected alternatives. The ADRs record not just what was decided but what was considered and why it lost — exactly the context an assistant needs to avoid re-proposing a rejected design or breaking a deliberate trade-off.

High branch coverage as a safety harness. >95% coverage with eager container validation means an AI-generated change that breaks a contract fails fast and loudly — at test time or at init(), not in production.

Tooling that closes the loop

  • pico-skills — installable skills for Claude Code and OpenAI Codex: /add-component, /add-controller, /add-repository, /add-actuator, and the pico-conventions background skill that teaches the assistant the whole ecosystem's API surface. One command:
curl -sL https://raw.githubusercontent.com/dperezcabrera/pico-skills/main/install.sh | bash

  • pico-initializer — scaffolds a ready-to-run project (web UI or CLI) that includes a tailored CLAUDE.md and an install-skills.sh, so a generated project is AI-maintainable from its first commit.

  • Zero-config plugins — modules like pico-actuator wire themselves through entry points and build from defaults when unconfigured. Less integration code means less code an assistant (or you) has to reason about to make a safe change.

Why it matters

The scarce resource in AI-assisted development is not writing code — it is verification and context. This ecosystem invests exactly there: contracts validated eagerly at startup, invariants written where assistants read them, uniform structure that makes review mechanical, and tests dense enough to catch a wrong change before a human ever sees it.

If you maintain your projects with an AI assistant, a pico app is built to be one of the easiest codebases it will ever work on.