The Agent Skills Directory

[COMMAND_EXECUTION]: The skill discovers and executes build commands defined in local project configuration files (e.g., package.json). While this is a core requirement for static site deployment, it involves executing commands found within the source code of the project being deployed.
[INDIRECT_PROMPT_INJECTION]: The skill reads and processes the contents of local files (HTML, CSS, JS, and config) to identify dependencies and rewrite assets during the staging process. This creates an attack surface where malicious code within a project could attempt to influence agent behavior.
Ingestion points: Local project files, including HTML, CSS, JavaScript, and configuration files.
Boundary markers: No explicit instruction delimiters are specified for the content being processed.
Capability inventory: Execution of build commands, file system modification (staging), and network upload via a dedicated tool.
Sanitization: Restricts all file access to the current workspace and validates that targets are regular files, rejecting symlinks and control characters.
[DATA_EXFILTRATION]: The skill's primary function is to upload local project data to public URLs and private cloud storage.
Evidence: Steps 5 and 6 detail the process of uploading the 'source_path' and 'path'.
Mitigation: The instructions explicitly require the agent to warn the user about public access and to obtain manual confirmation that the source contains no secrets before initiating the upload.

deploy-website