The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes system utilities such as git for repository history analysis and ffmpeg for video encoding through the subprocess module.
[EXTERNAL_DOWNLOADS]: The build-video.py script includes a setup routine to install the manim animation library from the official Python Package Index (PyPI) into a local virtual environment.
[DATA_EXFILTRATION]: Narration scripts are sent to the ElevenLabs API (api.elevenlabs.io) to generate audio. The skill documentation notes that this text describes code that may be private.
[DYNAMIC_EXECUTION]: Video explainers are rendered by executing Python code within scene.py using the Manim rendering engine at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill summarizes information from Pull Request discussions and repository files, which could allow malicious instructions embedded in PR comments to influence the agent's summary.
Ingestion points: Reads Pull Request bodies and comments (via gh pr view), commit history, and codebase files.
Boundary markers: The skill does not define specific delimiters to separate untrusted repo data from the summarizing instructions.
Capability inventory: File system writes, command execution (git, ffmpeg, manim), and network access (ElevenLabs).
Sanitization: No explicit filtering or sanitization is applied to the content retrieved from Pull Requests or commit logs.

understand