README

Expert patterns for modernizing PHP applications to PHP 8.x with type safety, PSR / PER-CS compliance, static-analysis tooling, and Symfony-grade architecture. The skill ships an agent contract — a small set of executable tools that emit machine-readable verification reports, archetype profiles, and fix-loop transcripts.

Compatibility

This is an Agent Skill following the open standard originally developed by Anthropic and released for cross-platform use.

Supported platforms:

• Claude Code (Anthropic)
• Cursor
• GitHub Copilot
• Other skills-compatible AI agents

Skills are portable packages of procedural knowledge that work across any AI agent supporting the Agent Skills specification.

Features

• Agent contract: a Python verifier (verify_php_project.py), orchestrator (modernize_loop.py), and cheap profiler (introspect.py). stable structured output (JSON with schema_version 1.0.0, SARIF 2.1.0, or JUnit XML), archetype detection, and machine-readable agent_actions[] recommendations downstream agents can act on without re-reading the prose references.
• Hard guardrails: five binding refusal cases enforced in SKILL.md (no readonly on Doctrine entities, no Rector without --dry-run, baseline shrink-not-delete, no blanket final on mock targets, no editing generated files).
• PHP 8.x feature coverage: 8.0–8.3 baseline, dedicated 8.4 reference (released 2024-11: property hooks, asymmetric visibility, lazy objects, array_find / array_any / array_all), and dedicated 8.5 reference (released 2025-11: pipe |>, array_first / array_last, #[\NoDiscard]).
• Static-analysis stack: PHPStan (level 9+, level 10 recommended), PHPat (architecture testing), Rector (automated refactoring), PHP-CS-Fixer (@PER-CS), Infection (mutation testing in PR-diff mode), composer audit.
• Type-safety patterns: DTOs and Value Objects over arrays, generic collection typing via PHPDoc, strict typing everywhere, immutability boundaries (readonly vs. property hooks vs. classic mutation).
• Framework edges: dedicated references for Doctrine 2.x/3.x edges, API Platform resource separation, and PSR-15 middleware architecture.
• PSR / PER-CS compliance: PSR-1, 3, 4, 6, 7, 11, 13, 14, 15, 16, 17, 18, 20 plus PER Coding Style. PSR-12 superseded by PER-CS.
• Project-archetype detection: generic-composer, typo3-extension, symfony-app, monorepo-package. The verifier and orchestrator branch on archetype.
• Regression suite: synthetic fixtures plus golden-output snapshots guard the verifier from contract drift.

Agent Contract

The skill follows a four-step loop. Cheap commands first; expensive analysis only when needed.

1. Discover — read SKILL.md, then profile the project.
2. Drill — narrow to one finding at a time when more than three things fail.
3. Apply — preview every fix in --mode dry-run before mutating files.
4. References — load references/<topic>.md only when the long-form rationale is needed.

One-line invocations

# Cheap first-touch profile (no subprocesses except `php --version`)
uv run skills/php-modernization/scripts/introspect.py --root .

# Full mechanical verification with agent_actions[]
uv run skills/php-modernization/scripts/verify_php_project.py --root . --format json --summary

# One checkpoint at a time
uv run skills/php-modernization/scripts/verify_php_project.py --root . --check PM-02

# Orchestrated fix preview (no mutations; use --confirm to apply)
uv run skills/php-modernization/scripts/modernize_loop.py --mode dry-run

uv is the recommended runner for the Python tools (it resolves the PEP 723 inline dependencies automatically). The Bash wrapper transparently falls back to python3 if uv is not installed.

Output formats

verify_php_project.py accepts --format json | sarif | junit. JSON is the canonical contract (see schemas/verification-result.schema.json). SARIF 2.1.0 plugs into the GitHub Security → Code scanning tab. JUnit XML is convenient for CI dashboards. Checkpoint IDs (PM-XX) are stable forever — never renumbered.

Hard Guardrails

The five guardrails below are binding. The agent must refuse rather than violate them.

• Never apply readonly to Doctrine entities or mapped-superclasses (embeddables are a nuanced case, ORM 3.x dependent — see references/doctrine-modernization-edges.md).
• Never run Rector without --dry-run first.
• Never raise the PHPStan level without regenerating + committing the baseline in the same change. Shrink, never delete.
• Never apply blanket final to mock targets or extension points without confirmation.
• Never edit @generated files or files under var/cache/, vendor/, node_modules/, .Build/.

The full table of references is in SKILL.md.

Installation

Marketplace (recommended)

Add the Netresearch marketplace once, then browse and install skills.

# Claude Code
/plugin marketplace add netresearch/claude-code-marketplace

npx (skills.sh)

Install with any Agent Skills-compatible agent:

npx skills add https://github.com/netresearch/php-modernization-skill --skill php-modernization

Composer (PHP projects)

composer require netresearch/php-modernization-skill

Requires netresearch/composer-agent-skill-plugin.

Download release

Download the latest release and extract to your agent's skills directory.

Git clone

git clone https://github.com/netresearch/php-modernization-skill.git

Usage

The skill auto-triggers on PHP modernization work: type-safety enforcement, PHPStan / Rector / PHP-CS-Fixer / PHPat setup, PSR / PER-CS compliance, Symfony patterns, version upgrades from 7.x or 8.0–8.3 baselines, and PHP 8.4 / 8.5 feature adoption.

Example queries:

• "Modernize this PHP class to PHP 8.4 with property hooks."
• "Add strict type safety with PHPStan level 10."
• "Set up PHPat architecture tests."
• "Configure Rector for a PHP 8.3 upgrade — dry-run only."
• "Make this HTTP client PSR-18 compliant."
• "Audit this Doctrine entity for incorrect readonly usage."

Verification & CI

Local

templates/composer-scripts.json is a drop-in scripts block for composer.json. Highlights:

• cs:fix, cs:check — PHP-CS-Fixer apply / dry-run
• phpstan — analysis without progress bar (CI-friendly)
• rector, rector:check — apply / dry-run
• phpat, audit — architecture tests + composer advisories
• skill:inspect, skill:verify — introspector / verifier shortcuts
• skill:fix, skill:qa — convenience pipelines

The script-name conventions (cs:fix, phpstan, rector) map onto checkpoints PM-13, PM-14, PM-15 — adopting them lights those checks green automatically. See templates/README.md for full guidance.

CI

templates/github-actions/php-modernization.yml is a copy-and-modify GitHub Actions workflow. It runs the verifier, optionally uploads SARIF to the Security → Code scanning tab, runs the orchestrator in --mode dry-run (with Infection diff-mode against the PR base), uploads artifacts, and posts a PR summary comment on failure. Adjust the php-version matrix to match your project's PHP requirement.

Project Archetypes

The verifier and orchestrator branch on archetype. Detection is pure (no subprocesses) and lives in scripts/_common.py::detect_archetype. Priority order:

A regression suite under fixtures/ covers each archetype plus a fully-modern/ positive control where every check passes. See fixtures/README.md for the snapshot workflow.

Repository Layout

php-modernization-skill/
├── skills/php-modernization/
│   ├── SKILL.md                       # Agent contract (router + guardrails)
│   ├── checkpoints.yaml               # PM-XX checkpoints
│   ├── references/                    # 19 lazy-loaded reference docs
│   ├── scripts/                       # verify_php_project.py + orchestrator + introspector
│   └── templates/                     # composer-scripts.json + GitHub Actions workflow
├── schemas/                           # JSON Schema 2020-12 (output contracts)
├── fixtures/                          # Synthetic-project regression suite
├── scripts/                           # test_fixtures.py + verify-harness.sh
├── docs/ARCHITECTURE.md               # Architecture overview
├── evals/                             # Skill evaluation suite
├── composer.json                      # Composer manifest
├── .claude-plugin/plugin.json         # Claude Code plugin manifest
└── sonar-project.properties           # SonarCloud config (excludes fixtures)

Required Static Analysis Tools

PSR / PER-CS Compliance

Source of truth: php-fig.org/psr and php-fig.org/per.

Type Safety: DTOs and Value Objects

Never pass or return raw arrays for structured data. Use typed objects.

Why: arrays lack runtime type safety, IDE autocompletion, and refactor-friendliness; PHPStan cannot verify array shapes across boundaries. Full patterns including Request DTOs, Command/Query DTOs, and Value Objects live in references/request-dtos.md. Immutability rules — when readonly applies, when property hooks fit better, when classic mutation is correct — are in references/immutability-boundaries.md.

Type Safety: Backed Enums

Never use string or integer constants, or arrays of literals, for fixed value sets. Use PHP 8.1+ backed enums.

Why: compile-time type checks, IDE autocompletion, exhaustive match() enforcement, and methods that encapsulate label / colour / validation logic. Full patterns are in references/php8-features.md.

Migration Checklist

• declare(strict_types=1) everywhere
• @PER-CS ruleset, no deprecated PHP-CS-Fixer aliases
• PHPStan ≥ 9 with treatPhpDocTypesAsCertain: false (level 10 for new projects)
• PHPat for layer boundaries
• Return + parameter types on every method
• DTOs over arrays; backed enums over constants
• PSR interfaces in type-hints (not implementations)
• #[Override] (8.3+), #[SensitiveParameter] (8.2+), typed constants (8.3+)
• readonly on DTOs / VOs / events only — see immutability boundaries
• Property hooks (8.4) for validated mutable state
• array_find / array_any / array_all over manual loops (8.4)
• Pipe |> for transform pipelines (8.5)

Related Skills

• security-audit-skill — security patterns for PHP applications.
• typo3-conformance and typo3-extension-upgrade — TYPO3-specific checkpoints (the IDs PM-26/27/28/33/35/36 were relocated there).
• typo3-testing-skill — PHPUnit patterns applicable to any PHP project.

License

This project uses split licensing:

• Code (scripts, workflows, configs): MIT
• Content (skill definitions, documentation, references): CC-BY-SA-4.0

See the individual license files for full terms.

Credits

Developed and maintained by Netresearch DTT GmbH.