The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external URLs and local files using a headless browser, creating a potential surface for indirect prompt injection.
Ingestion points: The scripts/layout-report.mjs and scripts/stress-content.mjs utilities accept arbitrary URLs or local file paths as input.
Boundary markers: The skill lacks explicit delimiters or instructions to the agent to disregard instructions embedded within the processed web content.
Capability inventory: The tools utilize Playwright to navigate to network locations, read local files (via file:// URLs), evaluate JavaScript in the browser context, and optionally write screenshots to the filesystem.
Sanitization: Content extracted from analyzed pages is included in reports without sanitization or filtering.
[DYNAMIC_EXECUTION]: The shared utility scripts/_shared/layout-probe.mjs dynamically loads the Playwright module from a path provided via the PLAYWRIGHT_MODULE environment variable, which constitutes dynamic loading from a computed path.
[EXTERNAL_DOWNLOADS]: The skill's dependencies include playwright and playwright-core, well-known packages that typically download browser binaries from external registries during installation.
[COMMAND_EXECUTION]: The skill includes Node.js scripts for layout analysis that are intended for direct execution in the developer or agent environment.

responsive-validation