The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill enables the execution of JavaScript/Playwright code strings on a remote server via the client.browsers.playwright.execute method. This server-side execution is a primary function of the SDK for high-reliability browser automation.
[INDIRECT_PROMPT_INJECTION]: The documentation in examples/examples.md demonstrates how to wrap Playwright execution as a tool for LLM agents. This creates a potential vulnerability surface for indirect prompt injection if untrusted data is processed by the agent.
Ingestion points: The code parameter in the execute_playwright tool template (defined in examples/examples.md) accepts arbitrary strings from the LLM agent's context.
Boundary markers: The provided tool examples do not implement delimiters or 'ignore embedded instructions' warnings to distinguish the agent's logic from potentially untrusted data.
Capability inventory: The skill allows for comprehensive browser interaction, including navigation, script execution, and access to page content within the Kernel browser environment.
Sanitization: The example templates do not perform sanitization or validation of the code strings before they are passed to the remote execution engine.
[COMMAND_EXECUTION]: The setup documentation specifies the use of shell commands for package installation (uv pip install) and mentions the kernel-cli for managing browser sessions via the command line.

kernel-python-sdk