The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to construct and execute docker run commands using the Bash tool. These commands incorporate file paths and targets provided by the user or discovered via Glob. If the agent does not properly sanitize these inputs, it may be vulnerable to shell command injection if a user provides a filename containing shell metacharacters (e.g., backticks, semicolons, or pipe symbols).- [DATA_EXFILTRATION]: The Docker command pattern utilizes a volume mount (-v "$(pwd)":/workspace), which grants the container full read and write access to the entire host directory where the command is executed. While standard for build processes, this broad access increases the risk of sensitive data exposure if the container execution is manipulated.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external JavaScript source files to perform syntax validation and display diff previews. This creates a surface for indirect prompt injection where malicious instructions hidden in the code could attempt to override the agent's behavior during the analysis phase.
Ingestion points: Source files located in Core/sites/admin-cabinet/assets/js/src/ and Extensions/*/public/assets/js/src/.
Boundary markers: None. The instructions do not specify the use of delimiters or 'ignore' instructions for the processed code content.
Capability inventory: The agent has access to Bash (shell execution), Read (file access), and Glob tools.
Sanitization: The skill does not mandate sanitization or escaping of the code content before the agent evaluates it for validation or reporting.- [REMOTE_CODE_EXECUTION]: The skill instructions include a reference to clone and build a Docker image from the MikoPBXUtils repository on GitHub. This involves downloading and executing build instructions from a remote source. As this targets the vendor's official repository, the reference itself is considered part of the primary skill purpose.

babel-compiler