The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and analyzes PHP source code and configuration files from the local project directory. This represents a potential surface for indirect prompt injection, as malicious instructions could be embedded in the processed code or comments.
Ingestion points: Processes files matching patterns like {src,Classes}/**/*.php and composer.json via verify_php_project.py and modernize_loop.py.
Boundary markers: The skill relies on structured JSON/SARIF output for tool reporting, which helps separate data from instructions, though the LLM review prompts (e.g., PM-20, PM-34) process raw code content.
Capability inventory: The skill can write to the file system (via modernize_loop.py --mode apply) and execute local shell commands (PHP binaries in vendor/bin).
Sanitization: Standard regex and script-based checks are used, but manual review of modernization changes is recommended by the skill's own guardrails.
[COMMAND_EXECUTION]: The Python scripts (modernize_loop.py, verify_php_project.py) execute local command-line tools such as php, composer, php-cs-fixer, phpstan, and rector. These are invoked using subprocess.run to perform the primary functions of project analysis and refactoring. The skill follows best practices by preferring direct binary execution over shell aliases to avoid argument forwarding issues.
[EXTERNAL_DOWNLOADS]: The skill includes a checkpoint and tool run for composer audit, which fetches vulnerability data from the official Composer/Packagist advisory database. This is a well-known service and the operation is intended for project security auditing.

php-modernization