playwright-cli
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFECREDENTIALS_UNSAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external, untrusted web pages, which is a known vector for indirect prompt injection attacks.
- Ingestion points: External data enters the agent context through
playwright-cli snapshot,playwright-cli find,playwright-cli eval, and the interactiveplaywright-cli show --annotatetool. - Boundary markers: There are no explicit markers or wrappers to distinguish between trusted instructions and untrusted web content in the snapshot output.
- Capability inventory: The skill allows for arbitrary JavaScript execution (
run-code,eval), network interception and modification (route), and local file system writes (state-save,screenshot,pdf). - Sanitization: While the instructions advise redacting sensitive values before reporting, there is no automated sanitization of web content before it is processed by the agent.
- [DYNAMIC_EXECUTION]: The skill provides
run-codeandevalcommands that facilitate the execution of arbitrary JavaScript code within the Playwright automation context or the browser's Document Object Model (DOM). This allows the agent to generate and run logic dynamically during a session. - [COMMAND_EXECUTION]: The documentation instructs users on how to execute shell commands for installation (
npm install -g @playwright/cli) and running test suites (npx playwright test), which involve direct command-line execution on the host system. - [DATA_EXFILTRATION]: Commands such as
playwright-cli state-save,playwright-cli console, andplaywright-cli requestsallow for the collection and saving of sensitive session data, including cookies, local storage, and network traffic. If misused, these could be used to harvest credentials or session tokens, though the skill provides best practices to mitigate this risk. - [CREDENTIALS_UNSAFE]: The skill explicitly manages authentication states via
state-saveandstate-load. While it provides guidance on using.gitignoreand redacting secrets, the mechanism involves storing plaintext or lightly structured session tokens in JSON files on the local filesystem.
Audit Metadata