The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies extensively on playwright-cli, npm, and npx to perform browser automation tasks and manage the environment.
[REMOTE_CODE_EXECUTION]: The run-code command allows for the execution of arbitrary JavaScript/Playwright code, and the --filename flag allows executing local script files. Additionally, the skill utilizes npx playwright test, which executes test scripts that may contain arbitrary code.
[DATA_EXFILTRATION]: The skill possesses multiple primitives for accessing and moving data:
playwright-cli state-save, cookie-list, and localstorage-list can access sensitive authentication tokens and session data.
playwright-cli upload and drop can be used to send local files to remote websites.
playwright-cli screenshot and pdf can capture visual data from the browser.
[INDIRECT_PROMPT_INJECTION]: The skill is highly susceptible to indirect prompt injection as it processes untrusted content from the web:
Ingestion points: Page content (HTML), page titles, URLs, console logs, and network request/response data.
Boundary markers: None identified in the instructions to prevent the agent from obeying instructions found on web pages.
Capability inventory: High-risk capabilities including run-code, eval, upload, and state-save are available to be triggered.
Sanitization: There is no evidence of sanitization or filtering for content retrieved from external websites before it is processed by the agent.
[DYNAMIC_EXECUTION]: The skill uses playwright-cli eval to execute JavaScript in the context of the web page and run-code to execute Playwright scripts at runtime.
[EXTERNAL_DOWNLOADS]: The skill instructions include commands to install global packages (npm install -g @playwright/cli@latest) and run uninstalled packages via npx, which involves downloading code from the npm registry at runtime.

playwright-cli