The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is centered around executing a wide variety of shell commands through the playwright-cli tool to drive a browser session, manage tabs, and handle network requests.
[DYNAMIC_EXECUTION]: The skill provides the eval and run-code commands, which allow for the execution of arbitrary JavaScript within the browser context. This is used for advanced automation tasks but represents a significant dynamic execution surface.
Evidence: Found in SKILL.md and references/running-code.md where JavaScript functions are passed as strings or loaded from files for execution via the Playwright API.
[DATA_EXFILTRATION]: The skill includes functionality to access, list, and export sensitive browser session data, such as cookies and local storage state.
Evidence: The state-save, cookie-list, localstorage-get, and sessionstorage-list commands described in SKILL.md and references/storage-state.md allow the retrieval of authentication state which could be exfiltrated if the agent is directed to transfer the saved state files.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external, untrusted websites through snapshots and search commands, creating a potential vector for indirect prompt injection.
Ingestion points: playwright-cli snapshot and playwright-cli find in SKILL.md read content from the live page into the agent's context.
Capability inventory: The agent has access to eval, run-code, and file-writing tools (state-save, screenshot).
Boundary markers: SKILL.md contains a warning that page-provided tool schemas (WebMCP) should be treated as untrusted data.
Sanitization: No explicit content sanitization or escaping mechanisms for external page data are defined in the instructions.

playwright-cli