The Agent Skills Directory

[DYNAMIC_EXECUTION]: The translations.md file provides a PHP script intended for translation verification that uses the require statement to dynamically load and execute local files from computed paths, such as require "$base/$loc/app.php". This pattern allows for the execution of arbitrary PHP code if the target files being loaded have been tampered with.
[COMMAND_EXECUTION]: The skill instructs the agent to run several local command-line tools and binaries, including vendor/bin/pint, vendor/bin/pest, npm run test:e2e, and php artisan. These commands are standard for project verification but represent a significant capability for local code execution.
[INDIRECT_PROMPT_INJECTION]: The skill processes project diffs and code changes to determine which verification gates to run. It lacks boundary markers or sanitization for this untrusted input. Because the skill triggers test runners and scripts that execute parts of the codebase, it is vulnerable to indirect prompt injection where malicious content in a diff could result in code execution during the verification process. 1. Ingestion points: The skill monitors and processes changes to various project files including PHP scripts, Blade templates, JavaScript, CSS, and translation files as described in SKILL.md. 2. Boundary markers: No specific delimiters or instructions are used to separate instructions from the data being processed. 3. Capability inventory: The agent is authorized to execute vendor/bin/pint, vendor/bin/pest, npm run, php artisan, and the PHP interpreter via php -r. 4. Sanitization: There is no evidence of content validation, escaping, or filtering for the external code and data being processed.

bagisto-change-verification