The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from git diffs and project files as part of its core review workflow, which creates an ingestion path for malicious instructions embedded in the analyzed code.
Ingestion points: The skill utilizes scripts/review-diff.sh to read git diffs and is instructed to perform deep reviews of application source files.
Capability inventory: The agent can execute the provided shell and PHP scripts, read project files, and generate code changes based on its analysis.
Boundary markers: No instructions are provided to the agent to treat analyzed code as potentially untrusted data or to use delimiters to separate code from instructions.
Sanitization: The skill does not implement filtering or sanitization for instructions contained within comments or code blocks of the files it reviews.
[COMMAND_EXECUTION]: The skill includes scripts/review-diff.sh and scripts/detect-laravel-context.php. The bash script executes git diff using arguments passed by the agent, which is a standard utility for developer tools but involves running shell commands on the local repository.

laravel-best-practices