The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses npx --package @playwright/cli in scripts/playwright_cli.sh to fetch and run the Playwright CLI. It also prompts the user to install @playwright/cli via npm if npx is unavailable. These actions target well-known package registries and developer tools.
[COMMAND_EXECUTION]: The script scripts/playwright_cli.sh executes the playwright-cli command with user-provided arguments. The skill also makes use of eval and run-code commands to execute JavaScript in the browser context, which is necessary for its stated purpose of browser automation.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes content from arbitrary external websites using tools like snapshot and eval. Malicious instructions embedded in a web page could be processed by the agent.
Ingestion points: snapshot commands and eval outputs in SKILL.md and references/cli.md which read data from remote URLs.
Capability inventory: Shell command execution via playwright_cli.sh, including browser controls (click, fill) and arbitrary JavaScript execution (eval, run-code).
Boundary markers: None present to distinguish extracted web content from agent instructions.
Sanitization: None present; raw element text and page data are returned directly to the agent context.

playwright