The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides utility functions that execute shell commands using execSync. Specifically, the getVideoDuration function in reference.md interpolates a file path directly into a shell string: execSync(ffprobe ... "${filePath}"). If the file path is influenced by untrusted input, this could lead to command injection.
[DYNAMIC_EXECUTION]: The skill's primary workflow involves generating and executing TypeScript/JavaScript scripts to control browser sessions. It utilizes page.evaluate and page.addStyleTag to inject and execute code within the browser context, and generates standalone script files (e.g., scripts/record-demo.ts) that are subsequently executed using npx.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from external websites, creating a surface for indirect prompt injection.
Ingestion points: The skill uses page.goto() to load content from arbitrary URLs defined by the user or potentially suggested by the agent.
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded content when the agent interacts with page elements or dismisses cookie banners.
Capability inventory: The skill possesses file system access (fs.mkdirSync, fs.renameSync) and shell execution capabilities (execSync, ffmpeg, ffprobe).
Sanitization: No sanitization is performed on page content (e.g., button text, form labels) before the agent interacts with it, which could lead to unintended actions if a page contains malicious instructions.

playwright-recording