playwright-cli

Warn

Audited by Gen Agent Trust Hub on Aug 24, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill makes extensive use of shell commands to control browser sessions, including playwright-cli, npx, and npm. These tools are authorized in the frontmatter and used for automation workflows.
  • [REMOTE_CODE_EXECUTION]: The skill provides run-code and eval commands which allow the agent to execute arbitrary JavaScript code within the browser context. This functionality is documented for tasks like handling geolocation, permissions, and complex UI workflows, but it introduces a high-impact vector for executing untrusted scripts if the agent is manipulated by malicious input.
  • [CREDENTIALS_UNSAFE]: The skill facilitates the retrieval and manipulation of sensitive authentication data. Commands like cookie-get and localstorage-get allow the agent to read session tokens. Furthermore, the state-save auth.json feature provides a mechanism to export the entire browser authentication state to a local file, which could lead to credential exposure if these files are not managed securely.
  • [EXTERNAL_DOWNLOADS]: The documentation includes instructions to install the Playwright CLI globally using npm install -g @playwright/cli@latest. This fetches the tool from the official public NPM registry.
  • [INDIRECT_PROMPT_INJECTION]: The skill enables the agent to ingest and act upon data from external, untrusted websites through snapshots and search functions.
  • Ingestion points: Browser snapshot, find, and eval commands in SKILL.md and references/element-attributes.md allow the agent to read page content.
  • Boundary markers: None identified; the instructions do not provide delimiters or warnings to ignore commands that may be embedded in the web content.
  • Capability inventory: The agent has the capability to write files (state-save, screenshot), perform network operations (browser navigation, route), and execute shell commands (Bash).
  • Sanitization: No explicit sanitization or content filtering is implemented for the data retrieved from external web pages before it is processed.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 24, 2026, 02:11 AM
Security Audit — agent-trust-hub — playwright-cli