The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool restricted to a specific whitelist of safe commands: git, mkdir, date, and wc. These are used to determine the files changed in a branch and to organize report output folders.
[DATA_EXFILTRATION]: The skill reads the entire project source code (e.g., app/**/*.php) using Read, Grep, and Glob. While the skill processes potentially sensitive application logic, it has no network permissions to exfiltrate data; all output is saved to the local project directory via the Write tool.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it performs 'deep reading' of untrusted source code provided by the user or project environment.
Ingestion points: Files within the app/, database/, tests/, routes/, config/, and resources/views/ directories.
Boundary markers: No delimiters are implemented to encapsulate the code content being analyzed.
Capability inventory: The skill can perform restricted Bash commands, write to the filesystem, and perform broad file system reads.
Sanitization: The skill does not sanitize or filter the content of the files it reads before processing them.

code-review-laravel