AI Agents

AI agents write Testo tests well — and just as well port existing tests to Testo from other frameworks.

The fastest path to your first test: install the package (see Getting Started) and ask an agent to write or migrate tests, pointing it at llms-full.txt (see below).

How agents read Testo's docs

Testo publishes its documentation in the llms.txt format — a single machine-readable file an AI fetches instead of crawling HTML pages.

File	Content	When to use
/llms.txt	Page list with short descriptions and best practices.	Quick overview of available documentation.
/llms-full.txt	Full text of all documentation pages.	When the agent needs complete context for code generation.

Skills for agents

Alongside llms.txt, Testo ships a set of AI skills — short instructions that an agent (Claude Code, Codex, and others) loads on demand when it spots a matching task. They cover writing tests, migrating from other frameworks, parameterization, benchmarks, coverage, and other scenarios. The full list lives in the skills/ folder.

Roughly: llms.txt tells the agent what the API offers, while a skill tells it when to use what and where the pitfalls are. When triggered, a skill sends the agent off to read llms.txt for the details — so documentation isn't duplicated and skills don't go stale alongside the API.

Installation via the Composer plugin

To avoid copying skills by hand from vendor/testo/testo/skills/ into wherever your agent looks for them (.claude/skills/, .cursor/skills/, etc.), there's a dedicated package — llm/skills — that lays out skills from trusted Composer dependencies for you:

composer require --dev llm/skills

Once installed, composer skills:update copies skills into .agents/skills/ (the path is configurable via extra.skills.target in composer.json). Testo is a trusted skill provider, so no extra configuration is needed. To make them visible to a specific agent, add its path to extra.skills.aliases — for example, .claude/skills for Claude Code:

{
    "extra": {
        "skills": {
            "aliases": [".claude/skills", ".cursor/skills"],
            "auto-sync": true
        }
    }
}

At those paths the plugin creates links (Windows junctions or POSIX symlinks) pointing at the main folder — the same set of skills is visible to every agent with no duplicated files.

Typical workflow

The same three steps work for new tests and for porting an existing suite:

1. Feed the context. Point the agent at llms-full.txt or paste the relevant section.
2. Describe the task. Provide the class or function you want to cover — or paste the old test you want to convert.
3. Review and run. Check the result and run vendor/bin/testo.

Example prompts

Writing tests from scratch:

Documentation: https://php-testo.github.io/llms-full.txt

Using the Testo PHP testing framework, write tests for the following class:

{paste your class here}

Migrating from another framework:

Documentation: https://php-testo.github.io/llms-full.txt

Rewrite the tests in the following folder to Testo. Keep the same assertions and data providers,
do not invent extra scenarios.

Tests folder: tests/Unit

Claude Code

Claude Code can fetch URLs during a conversation. Add an instruction to your project's CLAUDE.md:

When writing PHP tests, use the Testo framework.
Fetch the documentation before writing tests: https://php-testo.github.io/llms-full.txt

Claude Code will read this instruction and pull the documentation whenever it generates or migrates tests.

Cursor, Windsurf, and others

AI-powered code editors let you register external documentation as context. The typical approach is to add the documentation URL once and then reference it in chat.

• Cursor: Add the URL via Settings → Features → Docs → Add new doc. After indexing, reference it in chat with @Docs → select Testo.
• Windsurf: Paste the URL directly into a Cascade message.

For any IDE that doesn't support URL indexing, paste the contents of llms-full.txt into the chat as context.