The Agent Skills Directory

[COMMAND_EXECUTION]: The documentation demonstrates the use of the playwright.execute and process.exec functions to run JavaScript or shell scripts within remote browser VMs, which is a core architectural feature for reducing latency in automation tasks.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the @onkernel/sdk package and related tools, and it provides instructions for the kernel deploy command to upload local source code to the vendor's cloud platform.
[PROMPT_INJECTION]: The skill identifies an indirect prompt injection surface as it is designed to visit and process content from untrusted external websites.
Ingestion points: Data entering the agent context via page.goto, page.content, and page.extract in various example files.
Boundary markers: No explicit delimiters or boundary markers for untrusted web data are shown in the provided code snippets.
Capability inventory: The skill documentation describes extensive capabilities including file system access (browsers.fs.*), network operations (browsers.curl), and dynamic code execution within isolated browser environments.
Sanitization: The provided examples do not include explicit sanitization or validation of content retrieved from the web.

build-kernel-ts-sdk