The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides playwright-cli run-code and playwright-cli eval commands which execute arbitrary JavaScript strings within the browser's execution context. This allows for complex page manipulation and data access that could be abused if the agent processes untrusted inputs.
[CREDENTIALS_UNSAFE]: The skill includes dedicated commands for managing authentication data, including playwright-cli cookie-get, playwright-cli state-save, and playwright-cli localstorage-get. These can be used to extract sensitive session tokens and authentication states from the browser.
[DATA_EXPOSURE_AND_EXFILTRATION]: Instructions in references/running-code.md demonstrate how to read the system clipboard using navigator.clipboard.readText() within a run-code block, which could lead to the exposure of sensitive information stored in the user's clipboard.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process content from arbitrary websites via playwright-cli snapshot. It lacks explicit boundary markers or instructions to ignore malicious commands embedded in the processed web pages, creating a surface for indirect prompt injection. Ingestion points include playwright-cli goto (SKILL.md); Capability inventory includes run-code, eval, and cookie-get (SKILL.md, references/storage-state.md).
[REMOTE_CODE_EXECUTION]: The skill provides instructions for installing the Playwright CLI using npm install -g @playwright/cli@latest and running tests via npx playwright test. While these target well-known package registries, they involve the download and execution of external code.

playwright-cli