The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands constructed from task variables, such as {dev_server_command} and {task_id}. This pattern allows for command injection if these variables contain shell metacharacters and are sourced from untrusted task definitions.\n- [DYNAMIC_EXECUTION]: TypeScript test scripts (.spec.ts) and Playwright configuration files are generated at runtime by interpolating task-specific strings into templates. These scripts are subsequently executed, presenting a risk of arbitrary code execution if input strings like task titles or natural language steps are maliciously crafted to break out of string literals.\n- [INDIRECT_PROMPT_INJECTION]:\n
Ingestion points: The skill reads browser_acceptance JSON configuration (steps, name) and task metadata (task_id, task_title).\n
Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands during the interpolation of these fields into script templates or shell commands.\n
Capability inventory: The skill can perform file system writes (generating tests), shell execution (npx, kill, npm), and network operations (via Playwright's page.goto).\n
Sanitization: No sanitization or validation of user-provided strings is described before they are used in command lines or code generation.\n- [EXTERNAL_DOWNLOADS]: The skill checks for the presence of Playwright and suggests using npm init playwright@latest if missing, which is a standard procedure for this tool but involves downloading external code.

browser-test