The Agent Skills Directory

[COMMAND_EXECUTION]: The skill operates via the node clip.mjs command and uses execSync internally within clip.mjs to locate the global NPM root for dependency resolution.
[EXTERNAL_DOWNLOADS]: The script fetches article content and assets from user-provided URLs and requires the installation of the playwright-core package, a well-known browser automation library.
[DYNAMIC_EXECUTION]: The script utilizes page.evaluate to execute logic within the browser context and instructs the agent to generate small, site-specific Playwright scripts when standard flags are insufficient for complex page layouts.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external article URLs.
Ingestion points: Article content is ingested from remote URLs via the page.goto method in clip.mjs.
Boundary markers: The SKILL.md instructions explicitly mandate a separate reviewer agent to inspect the rendered clip and require the agent to confirm the article is fully accessible before processing.
Capability inventory: The skill can execute subprocesses (node), write files to the local system (PDF, PNG, and JSON metadata), and perform arbitrary network requests via the browser.
Sanitization: Instructions require the reviewer to verify all proposed CSS selectors against the live DOM and distinguish editorial content from advertisements to prevent data corruption or unintended deletions.

press-clip