The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow defined in SKILL.md uses an unsafe bash template: /usr/local/bin/python3 "$skill_dir/scripts/render_claude_typer.py" "<prompt>".
Directly interpolating user-supplied input into a shell command string allows for command injection if the input contains control characters like backticks, semicolons, or dollar signs.
[REMOTE_CODE_EXECUTION]: The script scripts/render_claude_typer.py renders content from https://www.laosunwendao.com using the Remotion CLI.
This involves executing JavaScript from an external, non-trusted source within a headless browser environment, which could be leveraged to exploit the local system or access sensitive files.
[PROMPT_INJECTION]: The inclusion of a --runner-prefix argument allows the agent to define the base command used for execution.
This high-privilege capability can be manipulated through prompt injection to trick the agent into executing arbitrary shell commands instead of the intended rendering tools.
[EXTERNAL_DOWNLOADS]: The skill downloads and executes packages from the npm registry at runtime and fetches composition data from a third-party domain.
It uses npx to download @remotion/cli and @remotion/tailwind-v4 from npm, and fetches assets from https://www.laosunwendao.com, which is not a recognized trusted vendor or well-known service.

claude-typer