The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides the ability to execute arbitrary JavaScript and Playwright code within a browser session via the -e flag and the execute MCP tool. This allows the agent to perform any action a user could in the browser environment.
[DATA_EXFILTRATION]: The skill connects to a running Chrome instance rather than a headless one, exposing all active user sessions, authentication cookies, and saved credentials to the agent. It also features a playwriter serve command that creates a WebSocket tunnel for remote access, which could be exploited for unauthorized access or silent harvesting of session data.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of a third-party Chrome extension (ID: jfeammnjpkecdekppnclgkkffahnhfhe) and a global NPM package (playwriter) from an external repository (remorses/playwriter).
[COMMAND_EXECUTION]: The skill relies on multiple CLI commands for browser interaction, session management, and setting up remote tunnels.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection by ingesting untrusted data from external websites.
Ingestion points: snapshot({ page }), getCleanHTML(), and getPageMarkdown() retrieve raw content from web pages.
Boundary markers: Absent; web content is processed directly by the agent without isolation delimiters.
Capability inventory: Execution of arbitrary code, manipulation of authenticated browser sessions, and data exfiltration via tunnels.
Sanitization: No evidence of sanitization or filtering of external instructions embedded in web content or metadata.

playwriter