The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to interact with and process content from external websites, which creates a surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context through web page navigation in scripts such as navigate.py and the custom automation templates provided in references/custom-scripts.md.
Boundary markers: There are no specialized delimiters or instructions to ignore embedded commands when the agent processes content retrieved from the browser.
Capability inventory: The skill can execute network requests, write files (screenshots, traces, and session state), and execute JavaScript within the browser context.
Sanitization: The provided templates do not include sanitization or validation of data retrieved from web pages before it is processed.
[COMMAND_EXECUTION]: The skill relies on executing local Python scripts using the uv package manager (e.g., uv run scripts/screenshot.py). While these scripts are intended to be part of the skill's infrastructure, they represent a mechanism for executing subprocesses on the host machine.
[DYNAMIC_EXECUTION]: The skill documentation and templates utilize Playwright's evaluate and evaluate_handle APIs, which allow the execution of arbitrary JavaScript within the context of a web page. This is a core feature for web automation but allows for the execution of logic determined at runtime.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for the user to install browser binaries using the Playwright CLI (playwright install chromium). This involves downloading large executable binaries from official, well-known Playwright repositories.

playwright