The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the user to execute various commands for installation and execution, including npx playwright install, pip install playwright, and running scripts via npx tsx or python. It also includes a shell script scripts/install-browsers.sh that automates browser binary installation.
[PRIVILEGE_ESCALATION]: The troubleshooting section for Ubuntu 25.10 suggests using sudo apt-get install to install system libraries (libicu72, libxml2). Additionally, the Docker deployment section recommends the --cap-add=SYS_ADMIN flag for local development to enable the Chromium sandbox.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download of browser binaries (~400MB) from Microsoft's infrastructure during the playwright install step. It also references external services like 2captcha.com and anti-captcha.com for solving CAPTCHAs and suggests using residential proxies.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from the web, creating a surface for indirect prompt injection.
Ingestion points: Web content is retrieved in templates/basic-scrape.ts using page.$$eval and page.textContent, and in templates/stealth-mode.ts using page.textContent('body').
Boundary markers: None identified in the provided templates to delimit scraped content from instructions.
Capability inventory: The skill can launch browser processes, perform network requests to any URL, and write files to the local system (e.g., session.json, screenshots, PDFs).
Sanitization: The templates do not demonstrate explicit sanitization or filtering of the scraped web content before it is logged or returned.
[DATA_EXFILTRATION]: The skill includes patterns for authenticated scraping in templates/authenticated-session.ts which saves browser cookies to a local file named session.json. While this is a standard persistence mechanism for automation, it involves the handling of sensitive session data.

playwright-local