The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The script scripts/check-conn.sh accesses the user's sensitive local configuration file ~/.claude.json to extract the PLAYWRIGHT_MCP_EXTENSION_TOKEN used for authenticating with the Playwright MCP extension.
[EXTERNAL_DOWNLOADS]: The connectivity check script scripts/check-conn.sh executes npx -y @playwright/mcp@latest, which fetches and runs the latest version of the Playwright MCP server package from the public npm registry at runtime.
[COMMAND_EXECUTION]: The skill utilizes shell scripts and instructs the agent to execute system-level PowerShell commands on the host machine, including triggering system beeps ([console]::beep) and launching the Edge browser with specific administrative pages.
[DYNAMIC_EXECUTION]: The skill relies on the browser_run_code_unsafe tool to execute arbitrary JavaScript code within the browser context, specifically for simulating human mouse movements to bypass Cloudflare Turnstile verification.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process large amounts of untrusted data (up to 200k characters) from external web pages to detect anti-bot challenges, creating a surface where malicious web content could influence agent behavior or drive automated browser actions.
Ingestion points: browser_evaluate and page.content() calls in SKILL.md and scripts/turnstile-click.js ingest raw HTML and text from external sites.
Boundary markers: The skill does not implement explicit boundary markers or delimiters when processing web content for decision-making.
Capability inventory: The skill has access to extensive browser capabilities including browser_run_code_unsafe, browser_click, browser_type, and browser_navigate, as well as host-level command execution.
Sanitization: External content is converted to lowercase for matching but is not sanitized or escaped before being used to trigger automated script execution.

p-edge-browser