The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the execution of multiple command-line utilities, including git for version control, pnpm for building applications, and the GitHub CLI (gh) for attaching media to pull requests. These operations are core to the skill's functionality for creating and publishing demos.
[DYNAMIC_EXECUTION]: The provided scripts/encode_gif.py script uses subprocess.run to invoke ffmpeg and ffprobe. It dynamically constructs the filter strings for media processing using parameters like FPS, width, and color depth. Security is maintained by strictly validating these inputs as positive integers or floats through the argparse configuration.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data by reading existing Pull Request bodies, repository commit history, and browser UI states. While this introduces a surface for indirect instructions, the skill includes several mitigation steps:
Ingestion points: Reads PR bodies, git rev-parse output, and browser-tool screenshots.
Boundary markers: None explicitly present for PR content, but the agent is instructed to perform specific targeted edits.
Capability inventory: The skill can write files, execute subprocesses (ffmpeg, git, pnpm, gh), and perform network operations via the GitHub CLI.
Sanitization: The skill mandates visual inspection of the final GIF and manual verification of the PR body changes to ensure no sensitive data is leaked or unintended instructions followed.

record-browser-gif