The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to download browser binaries via the official Playwright command-line interface (npx playwright install). This is a standard requirement for the tool's functionality and targets official Microsoft-maintained sources. It also suggests the use of well-known third-party services for captcha solving and proxy management, which are standard for the scraping use case.
[COMMAND_EXECUTION]: The skill includes shell scripts and TypeScript templates designed to be executed via standard runners like npx tsx and the playwright CLI. These commands are transparently documented and strictly relate to setting up the environment or running the automation scripts.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it ingests untrusted data from external websites into the agent's context during scraping operations.
Ingestion points: Data is pulled from web pages using page.textContent() and page.evaluate() in scripts like templates/basic-scrape.ts and templates/stealth-mode.ts.
Boundary markers: The templates do not currently implement explicit boundary markers or 'ignore' instructions for the agent when processing scraped text.
Capability inventory: The skill has access to file writing (fs.writeFile), network requests, and subprocess execution (npx tsx), which could be targeted by injected instructions.
Sanitization: Scraped content is returned to the agent without specific sanitization or filtering.
[DYNAMIC_EXECUTION]: The skill utilizes page.evaluate() to execute JavaScript within the browser context. While this is the standard operational mode for Playwright to interact with the DOM, it involves the dynamic execution of code within a controlled browser instance.

playwright-local