The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/encode_gif.py script and its companion test suite scripts/test_encode_gif.py utilize subprocess.run() to execute external media processing tools like ffmpeg and ffprobe.
Evidence: subprocess.run(command, check=True) in scripts/encode_gif.py and multiple calls in scripts/test_encode_gif.py to run the encoder and ffmpeg.
The implementation follows security best practices by passing arguments as lists rather than strings, which significantly mitigates shell injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection because it ingests and processes content from live browser pages that could be controlled by an attacker.
Ingestion Points: The agent records video and takes screenshots of browser UI content (SKILL.md).
Capabilities: The skill has the ability to write to the filesystem, execute local scripts, and interact with the gh (GitHub) CLI to modify pull request bodies.
Boundary Markers: The instructions include explicit guidelines for the agent to use isolated browser contexts, fresh profiles, and to actively avoid capturing secrets or personal data.
Sanitization: The Python encoder includes a ffconcat_quote function to sanitize file paths used in media manifests.

record-browser-gif