The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides tools for executing arbitrary JavaScript code within the browser context.
Evidence: The playwright-cli run-code command in SKILL.md and references/running-code.md executes a user-provided string as an asynchronous function.
Evidence: The playwright-cli eval command in SKILL.md and references/element-attributes.md executes code in the DOM context of a target element.
[COMMAND_EXECUTION]: The skill instructions frequently utilize host-level command execution via Bash tools for installation, testing, and debugging.
Evidence: SKILL.md and references/playwright-tests.md instruct the use of npm install, npm run, and npx playwright test to manage the environment and execute test suites.
[DATA_EXFILTRATION]: The skill contains functionalities designed to access and export sensitive browser session data.
Evidence: playwright-cli state-save in SKILL.md and references/storage-state.md saves complete browser state, including cookies and localStorage, to local files like auth.json.
Evidence: Commands such as cookie-list, cookie-get, localstorage-list, and sessionstorage-list provide direct access to potentially sensitive tokens and session identifiers.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to browse and interact with external, untrusted web pages, creating a significant attack surface for indirect injections.
Ingestion points: playwright-cli goto navigates to arbitrary URLs; playwright-cli snapshot and playwright-cli eval ingest content from the target page into the agent context (detected in SKILL.md).
Boundary markers: Absent. There are no instructions to the agent regarding the separation of data from the target page and its own instructions.
Capability inventory: High capabilities including run-code (JavaScript execution), Bash tools (npm/npx package management), and file system access (state-save, screenshot).
Sanitization: Absent. The skill provides no mechanisms for sanitizing content read from the browser before processing.
[EXTERNAL_DOWNLOADS]: The documentation includes instructions for downloading and installing external software packages.
Evidence: SKILL.md suggests installing the CLI tool via npm install -g @playwright/cli@latest. This targets the official Playwright package registry.
[CREDENTIALS_UNSAFE]: While the documentation recommends against committing storage state files, it explicitly provides mechanisms for handling sensitive credentials in cleartext.
Evidence: references/storage-state.md describes the JSON format of auth.json, which contains cleartext cookie values and localStorage items. The cookie-set command allows passing session IDs directly as command-line arguments.

playwright-cli