The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes Bash to run playwright-cli, npm, and npx. It provides run-code and eval commands that execute arbitrary JavaScript within the browser context, granting the agent high flexibility in page interaction.
[DATA_EXFILTRATION]: The skill includes dedicated commands to extract sensitive browser data, such as cookie-get and localstorage-get. It also allows saving the entire browser storage state to a file via state-save, which the documentation notes can contain authentication tokens.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core function of reading and snapshotting external website content. Evidence Chain: 1. Ingestion points: Untrusted data enters the context via playwright-cli snapshot and goto (SKILL.md). 2. Boundary markers: There are no instructions or delimiters provided to prevent the agent from obeying instructions embedded in the web content. 3. Capability inventory: The agent has access to Bash and extensive browser-based network/file operations. 4. Sanitization: No evidence of content filtering or sanitization exists in the provided workflow.
[EXTERNAL_DOWNLOADS]: The skill instructions recommend the installation of the @playwright/cli package from the official NPM registry.

playwright-cli