# PHP Modernization Skill 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](https://agentskills.io) 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 1.1.0`, 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 stated in `SKILL.md` (no `readonly` on Doctrine entities, no Rector apply without a reviewed 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 PSR-12 superseded by PER-CS.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 six-step loop. Cheap commands first; expensive analysis only when needed. 1. **Discover** — read [`SKILL.md`](skills/php-modernization/SKILL.md), then profile the project and read its own commands. 2. **Drill** — narrow to one finding at a time when more than three things fail. 3. **Use the project's tools** — run Rector, PHPStan, the coding-standard tool, lint and the tests the project has; report a check that did not run as not run. 4. **Decide what the tools cannot** — behaviour that must stay, enum or open value set, `match` comparisons, `readonly` over mutable members. 5. **Apply in bounded steps** — preview one scope in `--mode dry-run`, review the diff, apply, then run the checks again. 6. **References** — load `references/.md` only when the long-form rationale is needed. ### One-line invocations ```bash # 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). Without `uv`, the Bash wrapper falls back to `python3` only when that is Python 3.11 or later. ### Output formats `verify_php_project.py` accepts `--format json | sarif | junit`. JSON is the canonical contract (see [`schemas/verification-result.schema.json`](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`](skills/php-modernization/references/doctrine-modernization-edges.md)). - Never apply Rector before a `--dry-run` of the same scope and a review of its diff. - 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`](skills/php-modernization/SKILL.md#reference-routing). ## Installation ### Marketplace (recommended) Add the [Netresearch marketplace](https://github.com/netresearch/claude-code-marketplace) once, then browse and install skills. ```bash # Claude Code /plugin marketplace add netresearch/claude-code-marketplace /plugin install php-modernization@netresearch-claude-code-marketplace ``` ### Without a marketplace Since Claude Code 2.1.157 a plugin directory under your personal skills directory loads on its own: ```bash mkdir -p ~/.claude/skills git clone https://github.com/netresearch/php-modernization-skill.git \ ~/.claude/skills/php-modernization ``` It loads as `php-modernization@skills-dir` on the next session. Update with `git -C ~/.claude/skills/php-modernization pull` and start a new session; remove it by deleting the directory. This route has no `claude plugin update`. ### npx ([skills.sh](https://skills.sh)) Install with any [Agent Skills](https://agentskills.io)-compatible agent: ```bash npx skills add https://github.com/netresearch/php-modernization-skill --skill php-modernization ``` ### Composer (PHP projects) ```bash composer require netresearch/php-modernization-skill ``` Requires [netresearch/composer-agent-skill-plugin](https://github.com/netresearch/composer-agent-skill-plugin). ### Download release Download the [latest release](https://github.com/netresearch/php-modernization-skill/releases/latest) and extract to your agent's skills directory. ### Git clone ```bash 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`](skills/php-modernization/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`](skills/php-modernization/templates/README.md) for full guidance. ### CI [`templates/github-actions/php-modernization.yml`](skills/php-modernization/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: | Archetype | Detection signal | | --- | --- | | `typo3-extension` | `ext_emconf.php` or `Configuration/Services.yaml` | | `symfony-app` | `bin/console` plus `config/bundles.php` | | `monorepo-package` | `packages//composer.json` for two or more children | | `generic-composer` | `composer.json` plus either `src/` or a non-empty PSR-4 `autoload` map | A regression suite under [`fixtures/`](fixtures/) covers each archetype plus a `fully-modern/` positive control where every check passes. See [`fixtures/README.md`](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/ # 26 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 ├── tests/ # test_scripts.py (behaviour tests) ├── docs/ # ARCHITECTURE.md, SECURITY-ASSURANCE.md, php-rules.md ├── 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 | Tool | Purpose | Requirement | | --- | --- | --- | | [PHPStan](https://phpstan.org/) | Type checking, bug detection | Level 9 minimum, level 10 recommended | | [PHPat](https://www.phpat.dev/) | Architecture testing | Required for projects with defined layer boundaries | | [Rector](https://getrector.com/) | Automated refactoring | Required for modernization (always dry-run first) | | [PHP-CS-Fixer](https://cs.symfony.com/) | Coding style | Required with `@PER-CS` | | [Infection](https://infection.github.io/) | Mutation testing | Required in PR-diff mode | ## PSR / PER-CS Compliance | Standard | Purpose | Notes | | --- | --- | --- | | PSR-1 | Basic Coding | Required | | PSR-4 | Autoloading | Required | | PER Coding Style | Coding style | Required (supersedes PSR-12) | | PSR-12 | Extended coding style | Superseded by PER-CS — do not use | | PSR-3 | Logger | When logging | | PSR-6 / PSR-16 | Cache | When caching | | PSR-7 / PSR-17 / PSR-18 | HTTP messages, factories, client | For HTTP clients | | PSR-11 | Container | For DI containers | | PSR-13 | Hypermedia links | For HATEOAS responses | | PSR-14 | Events | For event dispatching | | PSR-15 | HTTP middleware | See [`references/psr15-middleware-architecture.md`](skills/php-modernization/references/psr15-middleware-architecture.md) | | PSR-20 | Clock | For time-dependent code | Source of truth: [php-fig.org/psr](https://www.php-fig.org/psr/) and [php-fig.org/per](https://www.php-fig.org/per/). ## Type Safety: DTOs and Value Objects Never pass or return raw arrays for structured data. Use typed objects. | Instead of | Use | | --- | --- | | `array $userData` | `UserDTO $user` | | `array{email: string, name: string}` | `readonly class UserDTO` | | `array $config` | `readonly class Config` or Value Object | | `array $request` | `RequestDTO::fromRequest($request)` | | `return ['success' => true, 'data' => $x]` | `return new ResultDTO($x)` | 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`](skills/php-modernization/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`](skills/php-modernization/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. | Instead of | Use | | --- | --- | | `const STATUS_DRAFT = 'draft'` | `enum Status: string { case Draft = 'draft'; }` | | `['draft', 'published', 'archived']` | `Status::cases()` | | `string $status` parameter | `Status $status` parameter | | `switch ($status)` | `match ($status)` with enum | 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`](skills/php-modernization/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. ## Contributing Open issues and pull requests on [GitHub](https://github.com/netresearch/php-modernization-skill). The commands for working on this repository are listed in [AGENTS.md](AGENTS.md#commands). ### Tests Two stdlib-only Python suites test the shipped scripts. They need Python 3.11 or later, which `uv run --python 3.11` provides: - `uv run --python 3.11 python scripts/test_fixtures.py` runs the verifier against every synthetic project under [`fixtures/`](fixtures/) and diffs its normalized JSON against `expected/verifier.json`. A failure prints `FAIL` and a unified diff per fixture; `--update` regenerates the snapshots when a change to the output is intended (see [`fixtures/README.md`](fixtures/README.md)). - `uv run --python 3.11 python tests/test_scripts.py` runs the introspector, the verifier, the orchestrator and the Bash wrapper as subprocesses and checks their exit codes, output formats, dry-run flags, the `--confirm` requirement of `--mode apply`, and that the reported version equals `.claude-plugin/plugin.json`. A failure names the test and the assertion. CI runs both suites on every pull request and every push to `main` through [`.github/workflows/tests.yml`](.github/workflows/tests.yml). `pre-commit run --all-files` runs the hooks of [`.pre-commit-config.yaml`](.pre-commit-config.yaml) locally; Skill Validation runs the linters among them (the skill validator, markdownlint, yamllint, actionlint, Ruff, ShellCheck) in CI. ### Dependencies - **Python tools:** standard library only. Each script declares `requires-python = ">=3.11"` and `dependencies = []` in its PEP 723 block, so `uv run` installs nothing. `uv` runs them on Python 3.11 or later; when `uv` is missing, the Bash wrapper uses `python3` only if that is 3.11 or later, and otherwise stops with an install hint. - **Composer:** `composer.json` requires `netresearch/composer-agent-skill-plugin`, which installs the skill into a PHP project. No `composer.lock` is committed (`.gitignore`). - **Development tools:** the hooks in `.pre-commit-config.yaml` are pinned by `rev:`. Renovate ([`renovate.json`](renovate.json), `config:recommended` with the pre-commit manager enabled) proposes updates for them. The Composer requirement is `*` and resolves to the latest release at install time. - **CI:** the workflows call reusable workflows of `netresearch/.github`, `netresearch/skill-repo-skill` and `netresearch/typo3-ci-workflows`; those reusables pin the actions they use by commit SHA (for example `validate.yml` and `tests.yml` of `netresearch/skill-repo-skill`). - **PHP tools** such as PHPStan, Rector or PHP-CS-Fixer are not dependencies of this repository. The scripts run them from the analysed project's `vendor/bin` or `.Build/bin` when they are installed there. ## Governance and policies This repository follows the Netresearch organisation policies: - [Governance](https://github.com/netresearch/.github/blob/main/GOVERNANCE.md): ownership, roles, how decisions are made and disputes resolved, and continuity. - [Roadmap](https://github.com/netresearch/.github/blob/main/ROADMAP.md): planned and explicitly excluded work for the coming year. - [Handling of dependency and code analysis findings](https://github.com/netresearch/.github/blob/main/SECURITY.md#handling-of-dependency-and-code-analysis-findings): thresholds, deadlines and the exception process for dependency (SCA) and static analysis (SAST) findings. - [Secret management](https://github.com/netresearch/.github/blob/main/SECURITY.md#secret-management): how CI and release credentials are stored, accessed and rotated. - [Access roster](https://github.com/netresearch/.github/blob/main/docs/access-roster.md): who holds administrative access to this repository and the organisation. The security assurance case for this skill (threat model, trust boundaries, countermeasures and limits) is in [docs/SECURITY-ASSURANCE.md](docs/SECURITY-ASSURANCE.md). Checks that run on pull requests in this repository: - Every pull request: Skill Validation (`lint.yml`: skill structure, markdownlint, yamllint, actionlint, JSON syntax, ShellCheck, Ruff, checkpoint schema), Eval Validation (`eval-validate.yml`), Skill Tests (`tests.yml`: `scripts/test_fixtures.py` and `tests/test_scripts.py`), PR Quality Gates (`pr-quality.yml`: approves pull requests whose author has write access or higher) and Label PR (`labeler.yml`). - Every pull request, skipped unless Renovate or Dependabot opened it: Auto-merge dependency PRs (`auto-merge-deps.yml`). - Pull requests to `main` or `master`: `security.yml` with Betterleaks (secret scanning), zizmor (workflow static analysis), dependency review (fails on vulnerabilities of severity high or above), Composer Audit and Opengrep SAST (thresholds as in the organisation's [static analysis rule](https://github.com/netresearch/.github/blob/main/SECURITY.md#static-analysis-sast)), and CodeQL (actions, python); Harness Verification (`harness-verify.yml`) and Template Drift (`check-template-drift.yml`). - Configured outside the workflow files: SonarCloud Code Analysis, the DCO check and the CodeRabbit review status. Code scanning also reports each analyser's results as a check run of its own: `CodeQL`, `SonarCloud`, `zizmor`, `betterleaks` and `Opengrep OSS`. GitHub secret scanning with push protection is enabled as a repository setting. ## License This project uses split licensing: - **Code** (scripts, workflows, configs): [MIT](LICENSE-MIT) - **Content** (skill definitions, documentation, references): [CC-BY-SA-4.0](LICENSE-CC-BY-SA-4.0) See the individual license files for full terms. ## Credits Developed and maintained by [Netresearch DTT GmbH](https://www.netresearch.de/).