The Agent Skills Directory

EXTERNAL_DOWNLOADS (LOW): The skill documentation specifies the installation of the Chromium browser via playwright install. It also references CSS from fonts.googleapis.com. These are trusted sources according to the TRUST-SCOPE-RULE.
COMMAND_EXECUTION (MEDIUM): The skill utilizes the Bash tool for system-level setup and rendering operations, which is a powerful capability that could be abused if the agent is misled.
PROMPT_INJECTION (MEDIUM): The skill is vulnerable to Indirect Prompt Injection through its subtitle processing workflow.
Ingestion points: Untrusted data is ingested from user-provided .srt files and analyzed to generate video overlay content.
Boundary markers: Absent; there are no clear delimiters or instructions to ignore embedded commands within the subtitle text.
Capability inventory: The skill has access to Bash, Read, and Write tools, as well as JavaScript execution within a browser renderer.
Sanitization: The template bullet-points.html uses .innerHTML to render points extracted from the subtitle analysis, creating an XSS surface where malicious subtitle content could execute code in the Playwright environment to probe the local system.

video-wrapper