The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses child_process.spawn in scripts/cdp.mjs to launch a headless Chrome instance with specific debugging flags. It also utilizes child_process.execSync in scripts/marked-build.mjs to perform Git operations (status, checkout) and execute build commands (e.g., npm run build) to facilitate testing of different application versions.
[DYNAMIC_EXECUTION]: The core driver provides page.eval() and page.addInitScript() functions, which execute JavaScript strings within the browser context. This is the primary mechanism for interacting with the DOM and instrumenting the application during testing.
[INDIRECT_PROMPT_INJECTION]: The skill implements methods to read text and DOM data from the browser via page.text() and page.eval(). This creates a vulnerability surface for indirect prompt injection if the agent is instructed to navigate to and process content from an untrusted external website.
Ingestion points: Browser DOM content is ingested into the agent context via page.text() and page.eval() in scripts/cdp.mjs.
Boundary markers: The implementation does not include specific boundary markers or 'ignore' instructions for data retrieved from the browser DOM.
Capability inventory: The skill has the capability to spawn subprocesses (Chrome), read and write files (screenshots, browser profiles), and execute shell commands (git and npm build scripts via marked-build.mjs).
Sanitization: No sanitization or escaping of the retrieved DOM content is performed before returning it to the agent.

kw-uat-browser-check