The Agent Skills Directory

[PROMPT_INJECTION]: Indirect prompt injection surface identified. 1. Ingestion points: Pull Request titles, bodies, and file paths via gh pr view (SKILL.md). 2. Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the processing logic. 3. Capability inventory: Browser interaction and file upload via agent-browser, Pull Request metadata modification via gh pr edit, and local command execution via ffmpeg and gh. 4. Sanitization: The skill lacks formal sanitization of PR data, though it instructs the agent to manually escape strings for JavaScript evaluation.
[COMMAND_EXECUTION]: The skill executes multiple shell commands to manage local directories (mkdir, rm), process video frames (ffmpeg), and manage GitHub repository resources (gh).
[DATA_EXFILTRATION]: The skill is designed to upload local video files to GitHub's attachment infrastructure. This is a primary feature and occurs after a user-reviewed shot list is created and manual authentication is performed.

feature-video