The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands including npx playwright test for test execution and ffmpeg for video format conversion and frame extraction.
[DYNAMIC_EXECUTION]: The skill generates temporary Playwright test scripts (e2e/zz-<name>-video.spec.ts) at runtime and executes them to capture specific user flows as video recordings.
[INDIRECT_PROMPT_INJECTION]: The skill presents a potential surface for indirect prompt injection as it ingests and processes untrusted data from web application UIs during the recording and verification phase.
Ingestion points: Web application content and UI elements accessed via Playwright (SKILL.md).
Boundary markers: Instructions explicitly require the agent to spot-check extracted frames and confirm captions render correctly, providing a manual verification step.
Capability inventory: The skill can execute local testing and video processing commands and perform file operations within the user's local directory (~/Documents/updates/).
Sanitization: The caption injection mechanism uses textContent to prevent DOM-based script injection within the recorded session, although the visual content itself is not sanitized against prompt injection aimed at the vision model.

demo-video