playwright-cli
Pass
Audited by Gen Agent Trust Hub on Jun 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads and processes content from external websites (HTML, console logs, and accessibility trees) that may contain malicious instructions designed to influence the agent's behavior.
- Ingestion points: Web content is ingested via
playwright-cli snapshot,playwright-cli console,playwright-cli network, andplaywright-cli evalas documented incore-commands.md,tracing-and-debugging.md, andrunning-custom-code.md. - Boundary markers: No specific delimiters or instructions to ignore embedded commands are present in the documentation to prevent the agent from obeying instructions found on web pages.
- Capability inventory: The agent has the capability to execute custom JavaScript (
run-code), write files (screenshot,pdf,state-save), and access local system resources like the clipboard and storage. - Sanitization: No evidence of input validation, escaping, or sanitization of ingested web data before processing is present in the skill instructions.
- [COMMAND_EXECUTION]: The skill provides an interface to a local command-line tool,
playwright-cli, which allows the agent to perform complex browser operations. This includes therun-codesubcommand which executes arbitrary JavaScript within the browser context, as seen throughoutadvanced-workflows.mdandrunning-custom-code.md. - [DATA_EXFILTRATION]: The skill includes multiple commands that provide access to sensitive session data, including cookies, localStorage, and clipboard content. While these are necessary for the skill's purpose, they represent a surface for data exposure or exfiltration if the agent is directed to process them maliciously.
- Evidence: Commands such as
cookie-list,localstorage-get, andstate-save(instorage-and-auth.md) enable reading and writing browser state to the local filesystem. - [EXTERNAL_DOWNLOADS]: The skill instructions (in
core-commands.md) direct users to install browsers and additional components usingplaywright-cli install --skillsandplaywright-cli install-browser.
Audit Metadata