The Agent Skills Directory

[Privilege Escalation] (HIGH): The script scripts/install-xdebug.sh uses sudo to perform system-level tasks including package installation via apt-get, extension installation via pecl, and writing configuration files to protected PHP directory structures. Modifying system-wide PHP configurations can impact other applications on the host.
[Remote Code Execution] (HIGH): The scripts/install-xdebug.sh file contains a pattern to download and pipe a script directly to bash (curl -fsSL ... | bash) for the Homebrew installer. While the source domain (GitHub) is within the [TRUST-SCOPE-RULE], the method of execution is a high-severity pattern.
[Command Execution] (MEDIUM): The skill facilitates the execution of arbitrary PHP code via Xdebug's eval command and various Laravel artisan commands. The scripts/setup-vscode.sh script automatically creates a tasks.json file which defines several shell tasks (php artisan migrate:fresh, etc.) that the agent can trigger.
[Indirect Prompt Injection] (HIGH): The skill is designed to ingest and analyze output from dd() and dump() calls within a Laravel application.
Ingestion points: capture_last_dump() and analyze_recent_dump() in SKILL.md read application data.
Boundary markers: No specific delimiters or instructions are provided to the agent to treat this data as untrusted.
Capability inventory: The skill possesses the ability to write files (via scripts/setup-vscode.sh) and execute system commands with elevated privileges (via scripts/install-xdebug.sh).
Sanitization: None detected. Malicious instructions embedded in application variables or database records that are subsequently 'dumped' could be interpreted by the agent as instructions.

laravel-debugging