The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill references the use of php artisan tinker and "executable-code tools" provided by the Laravel Boost server. These tools allow for the execution of arbitrary PHP code within the application's runtime environment. The skill attempts to mitigate this by suggesting these be limited to local/test environments unless authorized.
[COMMAND_EXECUTION]: The guidelines instruct the agent to use various CLI tools including php artisan (list, help, route:list, migrate, etc.), composer (show, install, update), and frontend package managers. This grants the agent the ability to execute a wide range of commands on the host system.
[INDIRECT_PROMPT_INJECTION]: The workflow requires the agent to ingest and analyze untrusted data from the repository, including composer.json, package.json, route files, and application bootstrap files. These files could contain malicious instructions designed to hijack the agent's logic.
Ingestion points: composer.json, composer.lock, package.json, bootstrap/app.php, bootstrap/providers.php, routes/web.php, routes/api.php, and files in docs/ or references/ directories.
Boundary markers: No specific delimiters or "ignore instructions" warnings are prescribed for the analysis of these repository files.
Capability inventory: File system modification (creating/updating controllers, models, migrations), shell command execution via Artisan/Composer, and database interaction.
Sanitization: While the skill warns against SQL interpolation, it lacks explicit sanitization or validation instructions for general data parsed from the repository's source code.

laravel-13-app-guidelines