The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses npx within the scripts/playwright_cli.sh wrapper to download and execute the @playwright/cli package from the npm registry. The instructions also suggest installing this package globally using npm install -g @playwright/cli@latest.
[COMMAND_EXECUTION]: The skill relies on executing shell commands via a wrapper script to control a browser. This includes operations like clicking elements, typing text, and capturing screenshots, which are executed in the user's terminal environment.
[INDIRECT_PROMPT_INJECTION]: The skill automates interaction with external websites and processes their content via the snapshot command. Because the content of these websites is external and untrusted, it could contain malicious instructions (e.g., hidden text or metadata) intended to influence the agent's behavior.
Ingestion points: Data enters the agent context through the open and snapshot commands which fetch and parse external DOM content.
Boundary markers: The instructions do not define specific delimiters or guardrails to distinguish between the skill's instructions and the content retrieved from the web.
Capability inventory: The agent has the ability to perform network requests, interact with the local file system (saving screenshots to output/playwright/), and execute arbitrary browser commands.
Sanitization: There is no evidence of sanitization or filtering of the HTML/DOM content before it is processed by the agent.

playwright