The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to launch a Playwright-managed Chromium binary from the Replit Nix store. It uses setsid and nohup to manage background daemon processes for the browser and the optional monitoring tool.
[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the playwright Python package using standard package managers (pip). This is a well-known library for browser automation.
[INDIRECT_PROMPT_INJECTION]: The skill enables an attack surface for indirect prompt injection as the browser can ingest untrusted content from the web.
Ingestion points: Browser instances launched via scripts/ensure_browser.sh or Playwright scripts in SKILL.md can load arbitrary URLs.
Boundary markers: None identified; external content is rendered directly in the browser environment.
Capability inventory: The agent has the capability to execute shell commands, perform file system operations in the workspace, and initiate network requests via the browser.
Sanitization: No content sanitization or instruction filtering is applied to the data processed by the browser.
[DATA_EXPOSURE]: The scripts/browser_monitor.py component starts a web server listening on all network interfaces (0.0.0.0). This server streams real-time screenshots of the browser activity. While intended for user debugging, this could expose sensitive information if the agent is browsing sensitive sites and the Replit port is publicly accessible.

replit-playwright-chromium